Skip to content

P
PaddleOCR

weixin_41840029 / PaddleOCR
与 Fork 源项目一致

Fork自 PaddlePaddle / PaddleOCR

通知 1
Star 1
Fork 0
P
PaddleOCR
未验证 提交 26219d5f 编写于 作者: S shaohua.zhang 提交者： GitHub
浏览文件
操作

Merge pull request #1 from PaddlePaddle/develop 

update-2020-7-17
上级 0e8a3417 311c5997
展开全部 隐藏空白更改
内联 并排
Showing with 58807 addition and 397 deletion
.clang_format.hook 0 → 100644
浏览文件 @ 26219d5f
#!/bin/bash
set -e
readonly VERSION="3.8"
version=$(clang-format -version)
if ! [[ $version == *"$VERSION"* ]]; then
echo "clang-format version check failed."
echo "a version contains '$VERSION' is needed, but get '$version'"
echo "you can install the right version, and make an soft-link to '\$PATH' env"
exit -1
fi
clang-format $@
.gitignore 0 → 100644
浏览文件 @ 26219d5f
# Byte-compiled / optimized / DLL files
__pycache__/
.ipynb_checkpoints/
*.py[cod]
*$py.class
# C extensions
*.so
inference/
inference_results/
output/
*.DS_Store
*.vs
*.user
*~
*.vscode
*.idea
*.log
.clang-format
.clang_format.hook
README.md
浏览文件 @ 26219d5f
[English](README_en.md) | 简体中文
## 简介 ## 简介
PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力使用者训练出更好的模型，并应用落地。 PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力使用者训练出更好的模型，并应用落地。
**直播预告：2020年7月21日晚8点B站直播，PaddleOCR开源大礼包全面解读，直播地址当天更新**
**近期更新**
- 2020.7.15 添加基于EasyEdge和Paddle-Lite的移动端DEMO，支持iOS和Android系统
- 2020.7.15 完善预测部署，添加基于C++预测引擎推理、服务化部署和端侧部署方案，以及超轻量级中文OCR模型预测耗时Benchmark
- 2020.7.15 整理OCR相关数据集、常用数据标注以及合成工具
- 2020.7.9 添加支持空格的识别模型，识别效果，预测及训练方式请参考快速开始和文本识别训练相关文档
- 2020.7.9 添加数据增强、学习率衰减策略,具体参考[配置文件](./doc/doc_ch/config.md)
- [more](./doc/doc_ch/update.md)
## 特性 ## 特性
- 超轻量级中文OCR，总模型仅8.6M - 超轻量级中文OCR模型，总模型仅8.6M
- 单模型支持中英文数字组合识别、竖排文本识别、长文本识别 - 单模型支持中英文数字组合识别、竖排文本识别、长文本识别
- 检测模型DB（4.1M）+识别模型CRNN（4.5M） - 检测模型DB（4.1M）+识别模型CRNN（4.5M）
- 实用通用中文OCR模型
- 多种预测推理部署方案，包括服务部署和端侧部署
- 多种文本检测训练算法，EAST、DB - 多种文本检测训练算法，EAST、DB
- 多种文本识别训练算法，Rosetta、CRNN、STAR-Net、RARE - 多种文本识别训练算法，Rosetta、CRNN、STAR-Net、RARE
- 可运行于Linux、Windows、MacOS等多种系统
## **超轻量级中文OCR体验** ## 快速体验
![](doc/imgs_results/11.jpg)
上图是超轻量级中文OCR模型效果展示，更多效果图请见文末[效果展示](#效果展示)。
#### 1.环境配置
请先参考[快速安装](./doc/installation.md)配置PaddleOCR运行环境。
#### 2.模型下载
``` <div align="center">
# 下载inference模型文件包 <img src="doc/imgs_results/11.jpg" width="800">
wget https://paddleocr.bj.bcebos.com/inference.tar </div>
# inference模型文件包解压
tar -xf inference.tar
```
#### 3.单张图像或者图像集合预测 上图是超轻量级中文OCR模型效果展示，更多效果图请见[效果展示页面](./doc/doc_ch/visualization.md)。
以下代码实现了文本检测、识别串联推理，在执行预测时，需要通过参数image_dir指定单张图像或者图像集合的路径、参数det_model_dir指定检测inference模型的路径和参数rec_model_dir指定识别inference模型的路径。可视化识别结果默认保存到 ./inference_results 文件夹里面。 - 超轻量级中文OCR在线体验地址：https://www.paddlepaddle.org.cn/hub/scene/ocr
- 移动端DEMO体验(基于EasyEdge和Paddle-Lite, 支持iOS和Android系统)：[安装包二维码获取地址](https://ai.baidu.com/easyedge/app/openSource?from=paddlelite)
``` Android手机也可以扫描下面二维码安装体验。
# 设置PYTHONPATH环境变量
export PYTHONPATH=.
# 预测image_dir指定的单张图像 <div align="center">
python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/" <img src="./doc/ocr-android-easyedge.png" width = "200" height = "200" />
</div>
# 预测image_dir指定的图像集合 - [**中文OCR模型快速使用**](./doc/doc_ch/quickstart.md)
python3 tools/infer/predict_system.py --image_dir="./doc/imgs/" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/"
# 如果想使用CPU进行预测，执行命令如下
python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/det/" --rec_model_dir="./inference/rec/" --use_gpu=False
```
更多的文本检测、识别串联推理使用方式请参考文档教程中[基于预测引擎推理](./doc/inference.md)。
## 中文OCR模型列表
|模型名称|模型简介|检测模型地址|识别模型地址|支持空格的识别模型地址|
|-|-|-|-|-|
|chinese_db_crnn_mobile|超轻量级中文OCR模型|[inference模型](https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db.tar)|[inference模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn.tar)|[inference模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_enhance_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_enhance.tar)
|chinese_db_crnn_server|通用中文OCR模型|[inference模型](https://paddleocr.bj.bcebos.com/ch_models/ch_det_r50_vd_db_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/ch_models/ch_det_r50_vd_db.tar)|[inference模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn.tar)|[inference模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_enhance_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_enhance.tar)
## 文档教程 ## 文档教程
- [快速安装](./doc/installation.md) - [快速安装](./doc/doc_ch/installation.md)
- [文本检测模型训练/评估/预测](./doc/detection.md)（持续更新中） - [中文OCR模型快速使用](./doc/doc_ch/quickstart.md)
- [文本识别模型训练/评估/预测](./doc/recognition.md)（持续更新中） - 算法介绍
- [基于预测引擎推理](./doc/inference.md)（持续更新中） - [文本检测](#文本检测算法)
- [文本识别](#文本识别算法)
## 文本检测算法 - [端到端OCR](#端到端OCR算法)
- 模型训练/评估
- [文本检测](./doc/doc_ch/detection.md)
- [文本识别](./doc/doc_ch/recognition.md)
- [yml参数配置文件介绍](./doc/doc_ch/config.md)
- [中文OCR训练预测技巧](./doc/doc_ch/tricks.md)
- 预测部署
- [基于Python预测引擎推理](./doc/doc_ch/inference.md)
- [基于C++预测引擎推理](./deploy/cpp_infer/readme.md)
- [服务化部署](./doc/doc_ch/serving.md)
- [端侧部署](./deploy/lite/readme.md)
- 模型量化压缩（coming soon）
- [Benchmark](./doc/doc_ch/benchmark.md)
- 数据集
- [通用中英文OCR数据集](./doc/doc_ch/datasets.md)
- [手写中文OCR数据集](./doc/doc_ch/handwritten_datasets.md)
- [垂类多语言OCR数据集](./doc/doc_ch/vertical_and_multilingual_datasets.md)
- [常用数据标注工具](./doc/doc_ch/data_annotation.md)
- [常用数据合成工具](./doc/doc_ch/data_synthesis.md)
- [FAQ](#FAQ)
- 效果展示
- [超轻量级中文OCR效果展示](#超轻量级中文OCR效果展示)
- [通用中文OCR效果展示](#通用中文OCR效果展示)
- [支持空格的中文OCR效果展示](#支持空格的中文OCR效果展示)
- [技术交流群](#欢迎加入PaddleOCR技术交流群)
- [参考文献](./doc/doc_ch/reference.md)
- [许可证书](#许可证书)
- [贡献代码](#贡献代码)
<a name="算法介绍"></a>
## 算法介绍
<a name="文本检测算法"></a>
### 1.文本检测算法
PaddleOCR开源的文本检测算法列表： PaddleOCR开源的文本检测算法列表：
- [x] EAST([paper](https://arxiv.org/abs/1704.03155)) - [x] EAST([paper](https://arxiv.org/abs/1704.03155))
- [x] DB([paper](https://arxiv.org/abs/1911.08947)) - [x] DB([paper](https://arxiv.org/abs/1911.08947))
- [ ] SAST([paper](https://arxiv.org/abs/1908.05498))(百度自研, comming soon) - [ ] SAST([paper](https://arxiv.org/abs/1908.05498))(百度自研, coming soon)
在ICDAR2015文本检测公开数据集上，算法效果如下： 在ICDAR2015文本检测公开数据集上，算法效果如下：
|模型|骨干网络|Hmean|下载链接| |模型|骨干网络|precision|recall|Hmean|下载链接|
|-|-|-|-|-|-|
|EAST|ResNet50_vd|88.18%|85.51%|86.82%|[下载链接](https://paddleocr.bj.bcebos.com/det_r50_vd_east.tar)|
|EAST|MobileNetV3|81.67%|79.83%|80.74%|[下载链接](https://paddleocr.bj.bcebos.com/det_mv3_east.tar)|
|DB|ResNet50_vd|83.79%|80.65%|82.19%|[下载链接](https://paddleocr.bj.bcebos.com/det_r50_vd_db.tar)|
|DB|MobileNetV3|75.92%|73.18%|74.53%|[下载链接](https://paddleocr.bj.bcebos.com/det_mv3_db.tar)|
使用[LSVT](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_ch/datasets.md#1icdar2019-lsvt)街景数据集共3w张数据，训练中文检测模型的相关配置和预训练文件如下：
|模型|骨干网络|配置文件|预训练模型|
|-|-|-|-| |-|-|-|-|
|EAST|ResNet50_vd|85.85%|[下载链接](https://paddleocr.bj.bcebos.com/det_r50_vd_east.tar)| |超轻量中文模型|MobileNetV3|det_mv3_db.yml|[下载链接](https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db.tar)|
|EAST|MobileNetV3|79.08%|[下载链接](https://paddleocr.bj.bcebos.com/det_mv3_east.tar)| |通用中文OCR模型|ResNet50_vd|det_r50_vd_db.yml|[下载链接](https://paddleocr.bj.bcebos.com/ch_models/ch_det_r50_vd_db.tar)|
|DB|ResNet50_vd|83.30%|[下载链接](https://paddleocr.bj.bcebos.com/det_r50_vd_db.tar)|
|DB|MobileNetV3|73.00%|[下载链接](https://paddleocr.bj.bcebos.com/det_mv3_db.tar)| * 注： 上述DB模型的训练和评估，需设置后处理参数box_thresh=0.6，unclip_ratio=1.5，使用不同数据集、不同模型训练，可调整这两个参数进行优化
PaddleOCR文本检测算法的训练和使用请参考文档教程中[文本检测模型训练/评估/预测](./doc/detection.md)。 PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训练/评估中的文本检测部分](./doc/doc_ch/detection.md)。
## 文本识别算法 <a name="文本识别算法"></a>
### 2.文本识别算法
PaddleOCR开源的文本识别算法列表： PaddleOCR开源的文本识别算法列表：
- [x] CRNN([paper](https://arxiv.org/abs/1507.05717)) - [x] CRNN([paper](https://arxiv.org/abs/1507.05717))
- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085)) - [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))
- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html)) - [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1)) - [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))
- [ ] SRN([paper](https://arxiv.org/abs/2003.12294))(百度自研, comming soon) - [ ] SRN([paper](https://arxiv.org/abs/2003.12294))(百度自研, coming soon)
参考[DTRB](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下： 参考[DTRB](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
...@@ -95,82 +138,76 @@ PaddleOCR开源的文本识别算法列表： ...@@ -95,82 +138,76 @@ PaddleOCR开源的文本识别算法列表：
|RARE|Resnet34_vd|84.90%|rec_r34_vd_tps_bilstm_attn|[下载链接](https://paddleocr.bj.bcebos.com/rec_r34_vd_tps_bilstm_attn.tar)| |RARE|Resnet34_vd|84.90%|rec_r34_vd_tps_bilstm_attn|[下载链接](https://paddleocr.bj.bcebos.com/rec_r34_vd_tps_bilstm_attn.tar)|
|RARE|MobileNetV3|83.32%|rec_mv3_tps_bilstm_attn|[下载链接](https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_attn.tar)| |RARE|MobileNetV3|83.32%|rec_mv3_tps_bilstm_attn|[下载链接](https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_attn.tar)|
PaddleOCR文本识别算法的训练和使用请参考文档教程中[文本识别模型训练/评估/预测](./doc/recognition.md)。 使用[LSVT](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_ch/datasets.md#1icdar2019-lsvt)街景数据集根据真值将图crop出来30w数据，进行位置校准。此外基于LSVT语料生成500w合成数据训练中文模型，相关配置和预训练文件如下：
|模型|骨干网络|配置文件|预训练模型|
|-|-|-|-|
|超轻量中文模型|MobileNetV3|rec_chinese_lite_train.yml|[下载链接](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn.tar)|
|通用中文OCR模型|Resnet34_vd|rec_chinese_common_train.yml|[下载链接](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn.tar)|
PaddleOCR文本识别算法的训练和使用请参考文档教程中[模型训练/评估中的文本识别部分](./doc/doc_ch/recognition.md)。
## 端到端OCR算法 <a name="端到端OCR算法"></a>
- [ ] [End2End-PSL](https://arxiv.org/abs/1909.07808)(百度自研, comming soon) ### 3.端到端OCR算法
- [ ] [End2End-PSL](https://arxiv.org/abs/1909.07808)(百度自研, coming soon)
<a name="效果展示"></a>
## 效果展示 ## 效果展示
![](doc/imgs_results/1.jpg)
![](doc/imgs_results/7.jpg)
![](doc/imgs_results/12.jpg)
![](doc/imgs_results/4.jpg)
![](doc/imgs_results/6.jpg)
![](doc/imgs_results/9.jpg)
![](doc/imgs_results/16.png)
![](doc/imgs_results/22.jpg)
## 参考文献
```
1. EAST:
@inproceedings{zhou2017east,
title={EAST: an efficient and accurate scene text detector},
author={Zhou, Xinyu and Yao, Cong and Wen, He and Wang, Yuzhi and Zhou, Shuchang and He, Weiran and Liang, Jiajun},
booktitle={Proceedings of the IEEE conference on Computer Vision and Pattern Recognition},
pages={5551--5560},
year={2017}
}
2. DB:
@article{liao2019real,
title={Real-time Scene Text Detection with Differentiable Binarization},
author={Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang},
journal={arXiv preprint arXiv:1911.08947},
year={2019}
}
3. DTRB:
@inproceedings{baek2019wrong,
title={What is wrong with scene text recognition model comparisons? dataset and model analysis},
author={Baek, Jeonghun and Kim, Geewook and Lee, Junyeop and Park, Sungrae and Han, Dongyoon and Yun, Sangdoo and Oh, Seong Joon and Lee, Hwalsuk},
booktitle={Proceedings of the IEEE International Conference on Computer Vision},
pages={4715--4723},
year={2019}
}
4. SAST:
@inproceedings{wang2019single,
title={A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning},
author={Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming},
booktitle={Proceedings of the 27th ACM International Conference on Multimedia},
pages={1277--1285},
year={2019}
}
5. SRN:
@article{yu2020towards,
title={Towards Accurate Scene Text Recognition with Semantic Reasoning Networks},
author={Yu, Deli and Li, Xuan and Zhang, Chengquan and Han, Junyu and Liu, Jingtuo and Ding, Errui},
journal={arXiv preprint arXiv:2003.12294},
year={2020}
}
6. end2end-psl:
@inproceedings{sun2019chinese,
title={Chinese Street View Text: Large-scale Chinese Text Reading with Partially Supervised Learning},
author={Sun, Yipeng and Liu, Jiaming and Liu, Wei and Han, Junyu and Ding, Errui and Liu, Jingtuo},
booktitle={Proceedings of the IEEE International Conference on Computer Vision},
pages={9086--9095},
year={2019}
}
```
<a name="超轻量级中文OCR效果展示"></a>
### 1.超轻量级中文OCR效果展示 [more](./doc/doc_ch/visualization.md)
<div align="center">
<img src="doc/imgs_results/1.jpg" width="800">
</div>
<a name="通用中文OCR效果展示"></a>
### 2.通用中文OCR效果展示 [more](./doc/doc_ch/visualization.md)
<div align="center">
<img src="doc/imgs_results/chinese_db_crnn_server/11.jpg" width="800">
</div>
<a name="支持空格的中文OCR效果展示"></a>
### 3.支持空格的中文OCR效果展示 [more](./doc/doc_ch/visualization.md)
<div align="center">
<img src="doc/imgs_results/chinese_db_crnn_server/en_paper.jpg" width="800">
</div>
<a name="FAQ"></a>
## FAQ
1. **转换attention识别模型时报错：KeyError: 'predict'**
问题已解，请更新到最新代码。
2. **关于推理速度**
图片中的文字较多时，预测时间会增，可以使用--rec_batch_num设置更小预测batch num，默认值为30，可以改为10或其他数值。
3. **服务部署与移动端部署**
预计6月中下旬会先后发布基于Serving的服务部署方案和基于Paddle Lite的移动端部署方案，欢迎持续关注。
4. **自研算法发布时间**
自研算法SAST、SRN、End2End-PSL都将在7-8月陆续发布，敬请期待。
[more](./doc/doc_ch/FAQ.md)
<a name="欢迎加入PaddleOCR技术交流群"></a>
## 欢迎加入PaddleOCR技术交流群
请扫描下面二维码，完成问卷填写，获取加群二维码和OCR方向的炼丹秘籍
<div align="center">
<img src="./doc/joinus.jpg" width = "200" height = "200" />
</div>
<a name="许可证书"></a>
## 许可证书 ## 许可证书
本项目的发布受<a href="https://github.com/PaddlePaddle/PaddleOCR/blob/master/LICENSE">Apache 2.0 license</a>许可认证。 本项目的发布受<a href="https://github.com/PaddlePaddle/PaddleOCR/blob/master/LICENSE">Apache 2.0 license</a>许可认证。
## 版本更新 <a name="贡献代码"></a>
## 贡献代码
## 如何贡献代码
我们非常欢迎你为PaddleOCR贡献代码，也十分感谢你的反馈。 我们非常欢迎你为PaddleOCR贡献代码，也十分感谢你的反馈。
- 非常感谢 [Khanh Tran](https://github.com/xxxpsyduck) 贡献了英文文档。
- 非常感谢 [zhangxin](https://github.com/ZhangXinNan)([Blog](https://blog.csdn.net/sdlypyzq)) 贡献新的可视化方式、添加.gitgnore、处理手动设置PYTHONPATH环境变量的问题
- 非常感谢 [lyl120117](https://github.com/lyl120117) 贡献打印网络结构的代码
- 非常感谢 [xiangyubo](https://github.com/xiangyubo) 贡献手写中文OCR数据集
- 非常感谢 [authorfu](https://github.com/authorfu) 贡献Android和[xiadeye](https://github.com/xiadeye) 贡献IOS的demo代码
README_en.md 0 → 100644
浏览文件 @ 26219d5f
English | [简体中文](README.md)
## INTRODUCTION
PaddleOCR aims to create a rich, leading, and practical OCR tools that help users train better models and apply them into practice.
**Recent updates**、
- 2020.7.9 Add recognition model to support space, [recognition result](#space Chinese OCR results). For more information: [Recognition](./doc/doc_ch/recognition.md) and [quickstart](./doc/doc_ch/quickstart.md)
- 2020.7.9 Add data auguments and learning rate decay strategies,please read [config](./doc/doc_en/config_en.md)
- 2020.6.8 Add [dataset](./doc/doc_en/datasets_en.md) and keep updating
- 2020.6.5 Support exporting `attention` model to `inference_model`
- 2020.6.5 Support separate prediction and recognition, output result score
- [more](./doc/doc_en/update_en.md)
## FEATURES
- Lightweight Chinese OCR model, total model size is only 8.6M
- Single model supports Chinese and English numbers combination recognition, vertical text recognition, long text recognition
- Detection model DB (4.1M) + recognition model CRNN (4.5M)
- Various text detection algorithms: EAST, DB
- Various text recognition algorithms: Rosetta, CRNN, STAR-Net, RARE
<a name="Supported-Chinese-model-list"></a>
### Supported Chinese models list:
|Model Name|Description |Detection Model link|Recognition Model link| Support for space Recognition Model link|
|-|-|-|-|-|
|chinese_db_crnn_mobile|lightweight Chinese OCR model|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db_infer.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db.tar)|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_infer.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn.tar)|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_enhance_infer.tar) / [pre-train model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_enhance.tar)
|chinese_db_crnn_server|General Chinese OCR model|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_det_r50_vd_db_infer.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/ch_models/ch_det_r50_vd_db.tar)|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_infer.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn.tar)|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_enhance_infer.tar) / [pre-train model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_enhance.tar)
For testing our Chinese OCR online：https://www.paddlepaddle.org.cn/hub/scene/ocr
**You can also quickly experience the lightweight Chinese OCR and General Chinese OCR models as follows:**
## **LIGHTWEIGHT CHINESE OCR AND GENERAL CHINESE OCR INFERENCE**
![](doc/imgs_results/11.jpg)
The picture above is the result of our lightweight Chinese OCR model. For more testing results, please see the end of the article [lightweight Chinese OCR results](#lightweight-Chinese-OCR-results) , [General Chinese OCR results](#General-Chinese-OCR-results) and [Support for space Recognition Model](#Space-Chinese-OCR-results).
#### 1. ENVIRONMENT CONFIGURATION
Please see [Quick installation](./doc/doc_en/installation_en.md)
#### 2. DOWNLOAD INFERENCE MODELS
#### (1) Download lightweight Chinese OCR models
*If wget is not installed in the windows system, you can copy the link to the browser to download the model. After model downloaded, unzip it and place it in the corresponding directory*
Copy the detection and recognition 'inference model' address in [Chinese model List](#Supported-Chinese-model-list), download and unpack:
```
mkdir inference && cd inference
# Download the detection part of the Chinese OCR and decompress it
wget {url/of/detection/inference_model} && tar xf {name/of/detection/inference_model/package}
# Download the recognition part of the Chinese OCR and decompress it
wget {url/of/recognition/inference_model} && tar xf {name/of/recognition/inference_model/package}
cd ..
```
Take lightweight Chinese OCR model as an example:
```
mkdir inference && cd inference
# Download the detection part of the lightweight Chinese OCR and decompress it
wget https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db_infer.tar && tar xf ch_det_mv3_db_infer.tar
# Download the recognition part of the lightweight Chinese OCR and decompress it
wget https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_infer.tar && tar xf ch_rec_mv3_crnn_infer.tar
# Download the space-recognized part of the lightweight Chinese OCR and decompress it
wget https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_enhance_infer.tar && tar xf ch_rec_mv3_crnn_enhance_infer.tar
cd ..
```
After the decompression is completed, the file structure should be as follows:
```
|-inference
|-ch_rec_mv3_crnn
|- model
|- params
|-ch_det_mv3_db
|- model
|- params
...
```
#### 3. SINGLE IMAGE AND BATCH PREDICTION
The following code implements text detection and recognition inference tandemly. When performing prediction, you need to specify the path of a single image or image folder through the parameter `image_dir`, the parameter `det_model_dir` specifies the path to detection model, and the parameter `rec_model_dir` specifies the path to the recognition model. The visual prediction results are saved to the `./inference_results` folder by default.
```bash
# Prediction on a single image by specifying image path to image_dir
python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/ch_det_mv3_db/" --rec_model_dir="./inference/ch_rec_mv3_crnn/"
# Prediction on a batch of images by specifying image folder path to image_dir
python3 tools/infer/predict_system.py --image_dir="./doc/imgs/" --det_model_dir="./inference/ch_det_mv3_db/" --rec_model_dir="./inference/ch_rec_mv3_crnn/"
# If you want to use CPU for prediction, you need to set the use_gpu parameter to False
python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/ch_det_mv3_db/" --rec_model_dir="./inference/ch_rec_mv3_crnn/" --use_gpu=False
```
To run inference of the Generic Chinese OCR model, follow these steps above to download the corresponding models and update the relevant parameters. Examples are as follows:
```
# Prediction on a single image by specifying image path to image_dir
python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/ch_det_r50_vd_db/" --rec_model_dir="./inference/ch_rec_r34_vd_crnn/"
```
To run inference of the space-Generic Chinese OCR model, follow these steps above to download the corresponding models and update the relevant parameters. Examples are as follows:
```
# Prediction on a single image by specifying image path to image_dir
python3 tools/infer/predict_system.py --image_dir="./doc/imgs_en/img_12.jpg" --det_model_dir="./inference/ch_det_r50_vd_db/" --rec_model_dir="./inference/ch_rec_r34_vd_crnn_enhance/"
```
For more text detection and recognition models, please refer to the document [Inference](./doc/doc_en/inference_en.md)
## DOCUMENTATION
- [Quick installation](./doc/doc_en/installation_en.md)
- [Text detection model training/evaluation/prediction](./doc/doc_en/detection_en.md)
- [Text recognition model training/evaluation/prediction](./doc/doc_en/recognition_en.md)
- [Inference](./doc/doc_en/inference_en.md)
- [Introduction of yml file](./doc/doc_en/config_en.md)
- [Dataset](./doc/doc_en/datasets_en.md)
- [FAQ]((#FAQ)
## TEXT DETECTION ALGORITHM
PaddleOCR open source text detection algorithms list:
- [x] EAST([paper](https://arxiv.org/abs/1704.03155))
- [x] DB([paper](https://arxiv.org/abs/1911.08947))
- [ ] SAST([paper](https://arxiv.org/abs/1908.05498))(Baidu Self-Research, comming soon)
On the ICDAR2015 dataset, the text detection result is as follows:
|Model|Backbone|precision|recall|Hmean|Download link|
|-|-|-|-|-|-|
|EAST|ResNet50_vd|88.18%|85.51%|86.82%|[Download link](https://paddleocr.bj.bcebos.com/det_r50_vd_east.tar)|
|EAST|MobileNetV3|81.67%|79.83%|80.74%|[Download link](https://paddleocr.bj.bcebos.com/det_mv3_east.tar)|
|DB|ResNet50_vd|83.79%|80.65%|82.19%|[Download link](https://paddleocr.bj.bcebos.com/det_r50_vd_db.tar)|
|DB|MobileNetV3|75.92%|73.18%|74.53%|[Download link](https://paddleocr.bj.bcebos.com/det_mv3_db.tar)|
For use of [LSVT](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_en/datasets_en.md#1-icdar2019-lsvt) street view dataset with a total of 3w training data，the related configuration and pre-trained models for Chinese detection task are as follows:
|Model|Backbone|Configuration file|Pre-trained model|
|-|-|-|-|
|lightweight Chinese model|MobileNetV3|det_mv3_db.yml|[Download link](https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db.tar)|
|General Chinese OCR model|ResNet50_vd|det_r50_vd_db.yml|[Download link](https://paddleocr.bj.bcebos.com/ch_models/ch_det_r50_vd_db.tar)|
* Note: For the training and evaluation of the above DB model, post-processing parameters box_thresh=0.6 and unclip_ratio=1.5 need to be set. If using different datasets and different models for training, these two parameters can be adjusted for better result.
For the training guide and use of PaddleOCR text detection algorithms, please refer to the document [Text detection model training/evaluation/prediction](./doc/doc_en/detection_en.md)
## TEXT RECOGNITION ALGORITHM
PaddleOCR open-source text recognition algorithms list:
- [x] CRNN([paper](https://arxiv.org/abs/1507.05717))
- [x] Rosetta([paper](https://arxiv.org/abs/1910.05085))
- [x] STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))
- [x] RARE([paper](https://arxiv.org/abs/1603.03915v1))
- [ ] SRN([paper](https://arxiv.org/abs/2003.12294))(Baidu Self-Research, comming soon)
Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow:
|Model|Backbone|Avg Accuracy|Module combination|Download link|
|-|-|-|-|-|
|Rosetta|Resnet34_vd|80.24%|rec_r34_vd_none_none_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_r34_vd_none_none_ctc.tar)|
|Rosetta|MobileNetV3|78.16%|rec_mv3_none_none_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_mv3_none_none_ctc.tar)|
|CRNN|Resnet34_vd|82.20%|rec_r34_vd_none_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_r34_vd_none_bilstm_ctc.tar)|
|CRNN|MobileNetV3|79.37%|rec_mv3_none_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_mv3_none_bilstm_ctc.tar)|
|STAR-Net|Resnet34_vd|83.93%|rec_r34_vd_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_r34_vd_tps_bilstm_ctc.tar)|
|STAR-Net|MobileNetV3|81.56%|rec_mv3_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_ctc.tar)|
|RARE|Resnet34_vd|84.90%|rec_r34_vd_tps_bilstm_attn|[Download link](https://paddleocr.bj.bcebos.com/rec_r34_vd_tps_bilstm_attn.tar)|
|RARE|MobileNetV3|83.32%|rec_mv3_tps_bilstm_attn|[Download link](https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_attn.tar)|
We use [LSVT](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_en/datasets_en.md#1-icdar2019-lsvt) dataset and cropout 30w traning data from original photos by using position groundtruth and make some calibration needed. In addition, based on the LSVT corpus, 500w synthetic data is generated to train the Chinese model. The related configuration and pre-trained models are as follows:
|Model|Backbone|Configuration file|Pre-trained model|
|-|-|-|-|
|lightweight Chinese model|MobileNetV3|rec_chinese_lite_train.yml|[Download link](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn.tar)|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_enhance_infer.tar) & [pre-trained model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_enhance.tar)|
|General Chinese OCR model|Resnet34_vd|rec_chinese_common_train.yml|[Download link](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn.tar)|[inference model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_enhance_infer.tar) & [pre-trained model](https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_enhance.tar)|
Please refer to the document for training guide and use of PaddleOCR text recognition algorithms [Text recognition model training/evaluation/prediction](./doc/doc_en/recognition_en.md)
## END-TO-END OCR ALGORITHM
- [ ] [End2End-PSL](https://arxiv.org/abs/1909.07808)(Baidu Self-Research, comming soon)
<a name="lightweight-Chinese-OCR-results"></a>
## LIGHTWEIGHT CHINESE OCR RESULTS
![](doc/imgs_results/1.jpg)
![](doc/imgs_results/7.jpg)
![](doc/imgs_results/12.jpg)
![](doc/imgs_results/4.jpg)
![](doc/imgs_results/6.jpg)
![](doc/imgs_results/9.jpg)
![](doc/imgs_results/16.png)
![](doc/imgs_results/22.jpg)
<a name="General-Chinese-OCR-results"></a>
## General Chinese OCR results
![](doc/imgs_results/chinese_db_crnn_server/11.jpg)
![](doc/imgs_results/chinese_db_crnn_server/2.jpg)
![](doc/imgs_results/chinese_db_crnn_server/8.jpg)
<a name="Space-Chinese-OCR-results"></a>
## space Chinese OCR results
### LIGHTWEIGHT CHINESE OCR RESULTS
![](doc/imgs_results/img_11.jpg)
### General Chinese OCR results
![](doc/imgs_results/chinese_db_crnn_server/en_paper.jpg)
<a name="FAQ"></a>
## FAQ
1. Error when using attention-based recognition model: KeyError: 'predict'
The inference of recognition model based on attention loss is still being debugged. For Chinese text recognition, it is recommended to choose the recognition model based on CTC loss first. In practice, it is also found that the recognition model based on attention loss is not as effective as the one based on CTC loss.
2. About inference speed
When there are a lot of texts in the picture, the prediction time will increase. You can use `--rec_batch_num` to set a smaller prediction batch size. The default value is 30, which can be changed to 10 or other values.
3. Service deployment and mobile deployment
It is expected that the service deployment based on Serving and the mobile deployment based on Paddle Lite will be released successively in mid-to-late June. Stay tuned for more updates.
4. Release time of self-developed algorithm
Baidu Self-developed algorithms such as SAST, SRN and end2end PSL will be released in June or July. Please be patient.
[more](./doc/doc_en/FAQ_en.md)
## WELCOME TO THE PaddleOCR TECHNICAL EXCHANGE GROUP
WeChat: paddlehelp, note OCR, our assistant will get you into the group~
<img src="./doc/paddlehelp.jpg" width = "200" height = "200" />
## REFERENCES
```
1. EAST:
@inproceedings{zhou2017east,
title={EAST: an efficient and accurate scene text detector},
author={Zhou, Xinyu and Yao, Cong and Wen, He and Wang, Yuzhi and Zhou, Shuchang and He, Weiran and Liang, Jiajun},
booktitle={Proceedings of the IEEE conference on Computer Vision and Pattern Recognition},
pages={5551--5560},
year={2017}
}
2. DB:
@article{liao2019real,
title={Real-time Scene Text Detection with Differentiable Binarization},
author={Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang},
journal={arXiv preprint arXiv:1911.08947},
year={2019}
}
3. DTRB:
@inproceedings{baek2019wrong,
title={What is wrong with scene text recognition model comparisons? dataset and model analysis},
author={Baek, Jeonghun and Kim, Geewook and Lee, Junyeop and Park, Sungrae and Han, Dongyoon and Yun, Sangdoo and Oh, Seong Joon and Lee, Hwalsuk},
booktitle={Proceedings of the IEEE International Conference on Computer Vision},
pages={4715--4723},
year={2019}
}
4. SAST:
@inproceedings{wang2019single,
title={A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning},
author={Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming},
booktitle={Proceedings of the 27th ACM International Conference on Multimedia},
pages={1277--1285},
year={2019}
}
5. SRN:
@article{yu2020towards,
title={Towards Accurate Scene Text Recognition with Semantic Reasoning Networks},
author={Yu, Deli and Li, Xuan and Zhang, Chengquan and Han, Junyu and Liu, Jingtuo and Ding, Errui},
journal={arXiv preprint arXiv:2003.12294},
year={2020}
}
6. end2end-psl:
@inproceedings{sun2019chinese,
title={Chinese Street View Text: Large-scale Chinese Text Reading with Partially Supervised Learning},
author={Sun, Yipeng and Liu, Jiaming and Liu, Wei and Han, Junyu and Ding, Errui and Liu, Jingtuo},
booktitle={Proceedings of the IEEE International Conference on Computer Vision},
pages={9086--9095},
year={2019}
}
```
## LICENSE
This project is released under <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/master/LICENSE">Apache 2.0 license</a>
## CONTRIBUTION
We welcome all the contributions to PaddleOCR and appreciate for your feedback very much.
- Many thanks to [Khanh Tran](https://github.com/xxxpsyduck) for contributing the English documentation.
- Many thanks to [zhangxin](https://github.com/ZhangXinNan) for contributing the new visualize function、add .gitgnore and discard set PYTHONPATH manually.
- Many thanks to [lyl120117](https://github.com/lyl120117) for contributing the code for printing the network structure.
configs/det/det_mv3_db.yml
浏览文件 @ 26219d5f
...@@ -6,7 +6,8 @@ Global: ...@@ -6,7 +6,8 @@ Global:
print_batch_step: 2 print_batch_step: 2
save_model_dir: ./output/det_db/ save_model_dir: ./output/det_db/
save_epoch_step: 200 save_epoch_step: 200
eval_batch_step: 5000 # evaluation is run every 5000 iterations after the 4000th iteration
eval_batch_step: [4000, 5000]
train_batch_size_per_card: 16 train_batch_size_per_card: 16
test_batch_size_per_card: 16 test_batch_size_per_card: 16
image_shape: [3, 640, 640] image_shape: [3, 640, 640]
...@@ -50,4 +51,4 @@ PostProcess: ...@@ -50,4 +51,4 @@ PostProcess:
thresh: 0.3 thresh: 0.3
box_thresh: 0.7 box_thresh: 0.7
max_candidates: 1000 max_candidates: 1000
unclip_ratio: 1.5 unclip_ratio: 2.0
\ No newline at end of file
configs/det/det_mv3_east.yml
浏览文件 @ 26219d5f
...@@ -6,7 +6,7 @@ Global: ...@@ -6,7 +6,7 @@ Global:
print_batch_step: 5 print_batch_step: 5
save_model_dir: ./output/det_east/ save_model_dir: ./output/det_east/
save_epoch_step: 200 save_epoch_step: 200
eval_batch_step: 5000 eval_batch_step: [5000, 5000]
train_batch_size_per_card: 16 train_batch_size_per_card: 16
test_batch_size_per_card: 16 test_batch_size_per_card: 16
image_shape: [3, 512, 512] image_shape: [3, 512, 512]
......
configs/det/det_r50_vd_db.yml
浏览文件 @ 26219d5f
...@@ -6,7 +6,7 @@ Global: ...@@ -6,7 +6,7 @@ Global:
print_batch_step: 2 print_batch_step: 2
save_model_dir: ./output/det_db/ save_model_dir: ./output/det_db/
save_epoch_step: 200 save_epoch_step: 200
eval_batch_step: 5000 eval_batch_step: [5000, 5000]
train_batch_size_per_card: 8 train_batch_size_per_card: 8
test_batch_size_per_card: 16 test_batch_size_per_card: 16
image_shape: [3, 640, 640] image_shape: [3, 640, 640]
......
configs/det/det_r50_vd_east.yml
浏览文件 @ 26219d5f
...@@ -6,7 +6,7 @@ Global: ...@@ -6,7 +6,7 @@ Global:
print_batch_step: 5 print_batch_step: 5
save_model_dir: ./output/det_east/ save_model_dir: ./output/det_east/
save_epoch_step: 200 save_epoch_step: 200
eval_batch_step: 5000 eval_batch_step: [5000, 5000]
train_batch_size_per_card: 8 train_batch_size_per_card: 8
test_batch_size_per_card: 16 test_batch_size_per_card: 16
image_shape: [3, 512, 512] image_shape: [3, 512, 512]
......
configs/rec/rec_benchmark_reader.yml
浏览文件 @ 26219d5f
...@@ -10,4 +10,3 @@ EvalReader: ...@@ -10,4 +10,3 @@ EvalReader:
TestReader: TestReader:
reader_function: ppocr.data.rec.dataset_traversal,LMDBReader reader_function: ppocr.data.rec.dataset_traversal,LMDBReader
lmdb_sets_dir: ./train_data/data_lmdb_release/evaluation/ lmdb_sets_dir: ./train_data/data_lmdb_release/evaluation/
infer_img: ./infer_img
\ No newline at end of file
configs/rec/rec_chinese_common_train.yml 0 → 100644
浏览文件 @ 26219d5f
Global:
algorithm: CRNN
use_gpu: true
epoch_num: 3000
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/rec_CRNN
save_epoch_step: 3
eval_batch_step: 2000
train_batch_size_per_card: 128
test_batch_size_per_card: 128
image_shape: [3, 32, 320]
max_text_length: 25
character_type: ch
character_dict_path: ./ppocr/utils/ppocr_keys_v1.txt
loss_type: ctc
distort: false
use_space_char: false
reader_yml: ./configs/rec/rec_chinese_reader.yml
pretrain_weights:
checkpoints:
save_inference_dir:
infer_img:
Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel
Backbone:
function: ppocr.modeling.backbones.rec_resnet_vd,ResNet
layers: 34
Head:
function: ppocr.modeling.heads.rec_ctc_head,CTCPredict
encoder_type: rnn
SeqRNN:
hidden_size: 256
Loss:
function: ppocr.modeling.losses.rec_ctc_loss,CTCLoss
Optimizer:
function: ppocr.optimizer,AdamDecay
base_lr: 0.0005
beta1: 0.9
beta2: 0.999
configs/rec/rec_chinese_lite_train.yml
浏览文件 @ 26219d5f
...@@ -14,10 +14,14 @@ Global: ...@@ -14,10 +14,14 @@ Global:
character_type: ch character_type: ch
character_dict_path: ./ppocr/utils/ppocr_keys_v1.txt character_dict_path: ./ppocr/utils/ppocr_keys_v1.txt
loss_type: ctc loss_type: ctc
distort: false
use_space_char: false
reader_yml: ./configs/rec/rec_chinese_reader.yml reader_yml: ./configs/rec/rec_chinese_reader.yml
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
configs/rec/rec_chinese_reader.yml
浏览文件 @ 26219d5f
...@@ -11,4 +11,3 @@ EvalReader: ...@@ -11,4 +11,3 @@ EvalReader:
TestReader: TestReader:
reader_function: ppocr.data.rec.dataset_traversal,SimpleReader reader_function: ppocr.data.rec.dataset_traversal,SimpleReader
infer_img: ./infer_img
configs/rec/rec_icdar15_reader.yml
浏览文件 @ 26219d5f
...@@ -11,4 +11,3 @@ EvalReader: ...@@ -11,4 +11,3 @@ EvalReader:
TestReader: TestReader:
reader_function: ppocr.data.rec.dataset_traversal,SimpleReader reader_function: ppocr.data.rec.dataset_traversal,SimpleReader
infer_img: ./infer_img
configs/rec/rec_icdar15_train.yml
浏览文件 @ 26219d5f
...@@ -13,10 +13,14 @@ Global: ...@@ -13,10 +13,14 @@ Global:
max_text_length: 25 max_text_length: 25
character_type: en character_type: en
loss_type: ctc loss_type: ctc
distort: true
debug: false
reader_yml: ./configs/rec/rec_icdar15_reader.yml reader_yml: ./configs/rec/rec_icdar15_reader.yml
pretrain_weights: ./pretrain_models/rec_mv3_none_bilstm_ctc/best_accuracy pretrain_weights: ./pretrain_models/rec_mv3_none_bilstm_ctc/best_accuracy
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
...@@ -39,3 +43,7 @@ Optimizer: ...@@ -39,3 +43,7 @@ Optimizer:
base_lr: 0.0005 base_lr: 0.0005
beta1: 0.9 beta1: 0.9
beta2: 0.999 beta2: 0.999
decay:
function: cosine_decay
step_each_epoch: 20
total_epoch: 1000
configs/rec/rec_mv3_none_bilstm_ctc.yml
浏览文件 @ 26219d5f
...@@ -17,6 +17,7 @@ Global: ...@@ -17,6 +17,7 @@ Global:
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
configs/rec/rec_mv3_none_none_ctc.yml
浏览文件 @ 26219d5f
...@@ -17,6 +17,7 @@ Global: ...@@ -17,6 +17,7 @@ Global:
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
configs/rec/rec_mv3_tps_bilstm_attn.yml
浏览文件 @ 26219d5f
...@@ -13,11 +13,14 @@ Global: ...@@ -13,11 +13,14 @@ Global:
max_text_length: 25 max_text_length: 25
character_type: en character_type: en
loss_type: attention loss_type: attention
tps: true
reader_yml: ./configs/rec/rec_benchmark_reader.yml reader_yml: ./configs/rec/rec_benchmark_reader.yml
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
configs/rec/rec_mv3_tps_bilstm_ctc.yml
浏览文件 @ 26219d5f
...@@ -13,10 +13,12 @@ Global: ...@@ -13,10 +13,12 @@ Global:
max_text_length: 25 max_text_length: 25
character_type: en character_type: en
loss_type: ctc loss_type: ctc
tps: true
reader_yml: ./configs/rec/rec_benchmark_reader.yml reader_yml: ./configs/rec/rec_benchmark_reader.yml
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
......
configs/rec/rec_r34_vd_none_bilstm_ctc.yml
浏览文件 @ 26219d5f
...@@ -17,7 +17,9 @@ Global: ...@@ -17,7 +17,9 @@ Global:
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
configs/rec/rec_r34_vd_none_none_ctc.yml
浏览文件 @ 26219d5f
...@@ -17,6 +17,7 @@ Global: ...@@ -17,6 +17,7 @@ Global:
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
configs/rec/rec_r34_vd_tps_bilstm_attn.yml
浏览文件 @ 26219d5f
...@@ -13,10 +13,13 @@ Global: ...@@ -13,10 +13,13 @@ Global:
max_text_length: 25 max_text_length: 25
character_type: en character_type: en
loss_type: attention loss_type: attention
tps: true
reader_yml: ./configs/rec/rec_benchmark_reader.yml reader_yml: ./configs/rec/rec_benchmark_reader.yml
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
configs/rec/rec_r34_vd_tps_bilstm_ctc.yml
浏览文件 @ 26219d5f
...@@ -13,10 +13,13 @@ Global: ...@@ -13,10 +13,13 @@ Global:
max_text_length: 25 max_text_length: 25
character_type: en character_type: en
loss_type: ctc loss_type: ctc
tps: true
reader_yml: ./configs/rec/rec_benchmark_reader.yml reader_yml: ./configs/rec/rec_benchmark_reader.yml
pretrain_weights: pretrain_weights:
checkpoints: checkpoints:
save_inference_dir: save_inference_dir:
infer_img:
Architecture: Architecture:
function: ppocr.modeling.architectures.rec_model,RecModel function: ppocr.modeling.architectures.rec_model,RecModel
......
deploy/android_demo/.gitignore 0 → 100644
浏览文件 @ 26219d5f
*.iml
.gradle
/local.properties
/.idea/*
.DS_Store
/build
/captures
.externalNativeBuild
deploy/android_demo/README.md 0 → 100644
浏览文件 @ 26219d5f
# 如何快速测试
### 1. 安装最新版本的Android Studio
可以从https://developer.android.com/studio下载。本Demo使用是4.0版本Android Studio编写。
### 2. 按照NDK 20 以上版本
Demo测试的时候使用的是NDK 20b版本，20版本以上均可以支持编译成功。
如果您是初学者，可以用以下方式安装和测试NDK编译环境。
点击 File -> New ->New Project， 新建 "Native C++" project
### 3. 导入项目
点击 File->New->Import Project...， 然后跟着Android Studio的引导导入
# 获得更多支持
前往[端计算模型生成平台EasyEdge](https://ai.baidu.com/easyedge/app/open_source_demo?referrerUrl=paddlelite)，获得更多开发支持：
- Demo APP：可使用手机扫码安装，方便手机端快速体验文字识别
- SDK：模型被封装为适配不同芯片硬件和操作系统SDK，包括完善的接口，方便进行二次开发
deploy/android_demo/app/build.gradle 0 → 100644
浏览文件 @ 26219d5f
import java.security.MessageDigest
apply plugin: 'com.android.application'
android {
compileSdkVersion 28
defaultConfig {
applicationId "com.baidu.paddle.lite.demo.ocr"
minSdkVersion 15
targetSdkVersion 28
versionCode 1
versionName "1.0"
testInstrumentationRunner "android.support.test.runner.AndroidJUnitRunner"
externalNativeBuild {
cmake {
cppFlags "-std=c++11 -frtti -fexceptions -Wno-format"
arguments '-DANDROID_PLATFORM=android-23', '-DANDROID_STL=c++_shared' ,"-DANDROID_ARM_NEON=TRUE"
}
}
ndk {
// abiFilters "arm64-v8a", "armeabi-v7a"
abiFilters "arm64-v8a", "armeabi-v7a"
ldLibs "jnigraphics"
}
}
buildTypes {
release {
minifyEnabled false
proguardFiles getDefaultProguardFile('proguard-android-optimize.txt'), 'proguard-rules.pro'
}
}
externalNativeBuild {
cmake {
path "src/main/cpp/CMakeLists.txt"
version "3.10.2"
}
}
}
dependencies {
implementation fileTree(include: ['*.jar'], dir: 'libs')
implementation 'com.android.support:appcompat-v7:28.0.0'
implementation 'com.android.support.constraint:constraint-layout:1.1.3'
implementation 'com.android.support:design:28.0.0'
testImplementation 'junit:junit:4.12'
androidTestImplementation 'com.android.support.test:runner:1.0.2'
androidTestImplementation 'com.android.support.test.espresso:espresso-core:3.0.2'
}
def archives = [
[
'src' : 'https://paddlelite-demo.bj.bcebos.com/libs/android/paddle_lite_libs_v2_6_1.tar.gz',
'dest': 'PaddleLite'
],
[
'src' : 'https://paddlelite-demo.bj.bcebos.com/libs/android/opencv-4.2.0-android-sdk.tar.gz',
'dest': 'OpenCV'
],
[
'src' : 'https://paddleocr.bj.bcebos.com/deploy/lite/ocr_v1_for_cpu.tar.gz',
'dest' : 'src/main/assets/models/ocr_v1_for_cpu'
]
]
task downloadAndExtractArchives(type: DefaultTask) {
doFirst {
println "Downloading and extracting archives including libs and models"
}
doLast {
// Prepare cache folder for archives
String cachePath = "cache"
if (!file("${cachePath}").exists()) {
mkdir "${cachePath}"
}
archives.eachWithIndex { archive, index ->
MessageDigest messageDigest = MessageDigest.getInstance('MD5')
messageDigest.update(archive.src.bytes)
String cacheName = new BigInteger(1, messageDigest.digest()).toString(32)
// Download the target archive if not exists
boolean copyFiles = !file("${archive.dest}").exists()
if (!file("${cachePath}/${cacheName}.tar.gz").exists()) {
ant.get(src: archive.src, dest: file("${cachePath}/${cacheName}.tar.gz"))
copyFiles = true; // force to copy files from the latest archive files
}
// Extract the target archive if its dest path does not exists
if (copyFiles) {
copy {
from tarTree("${cachePath}/${cacheName}.tar.gz")
into "${archive.dest}"
}
}
}
}
}
preBuild.dependsOn downloadAndExtractArchives
\ No newline at end of file
deploy/android_demo/app/proguard-rules.pro 0 → 100644
浏览文件 @ 26219d5f
# Add project specific ProGuard rules here.
# You can control the set of applied configuration files using the
# proguardFiles setting in build.gradle.
#
# For more details, see
# http://developer.android.com/guide/developing/tools/proguard.html
# If your project uses WebView with JS, uncomment the following
# and specify the fully qualified class name to the JavaScript interface
# class:
#-keepclassmembers class fqcn.of.javascript.interface.for.webview {
# public *;
#}
# Uncomment this to preserve the line number information for
# debugging stack traces.
#-keepattributes SourceFile,LineNumberTable
# If you keep the line number information, uncomment this to
# hide the original source file name.
#-renamesourcefileattribute SourceFile
deploy/android_demo/app/src/androidTest/java/com/baidu/paddle/lite/demo/ocr/ExampleInstrumentedTest.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import android.content.Context;
import android.support.test.InstrumentationRegistry;
import android.support.test.runner.AndroidJUnit4;
import org.junit.Test;
import org.junit.runner.RunWith;
import static org.junit.Assert.*;
/**
* Instrumented test, which will execute on an Android device.
*
* @see <a href="http://d.android.com/tools/testing">Testing documentation</a>
*/
@RunWith(AndroidJUnit4.class)
public class ExampleInstrumentedTest {
@Test
public void useAppContext() {
// Context of the app under test.
Context appContext = InstrumentationRegistry.getTargetContext();
assertEquals("com.baidu.paddle.lite.demo", appContext.getPackageName());
}
}
deploy/android_demo/app/src/main/AndroidManifest.xml 0 → 100644
浏览文件 @ 26219d5f
<?xml version="1.0" encoding="utf-8"?>
<manifest xmlns:android="http://schemas.android.com/apk/res/android"
package="com.baidu.paddle.lite.demo.ocr">
<uses-permission android:name="android.permission.WRITE_EXTERNAL_STORAGE"/>
<uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE"/>
<uses-permission android:name="android.permission.CAMERA"/>
<application
android:allowBackup="true"
android:icon="@mipmap/ic_launcher"
android:label="@string/app_name"
android:roundIcon="@mipmap/ic_launcher_round"
android:supportsRtl="true"
android:theme="@style/AppTheme">
<activity android:name="com.baidu.paddle.lite.demo.ocr.MainActivity">
<intent-filter>
<action android:name="android.intent.action.MAIN"/>
<category android:name="android.intent.category.LAUNCHER"/>
</intent-filter>
</activity>
<activity
android:name="com.baidu.paddle.lite.demo.ocr.SettingsActivity"
android:label="Settings">
</activity>
</application>
</manifest>
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/CMakeLists.txt 0 → 100644
浏览文件 @ 26219d5f
# For more information about using CMake with Android Studio, read the
# documentation: https://d.android.com/studio/projects/add-native-code.html
# Sets the minimum version of CMake required to build the native library.
cmake_minimum_required(VERSION 3.4.1)
# Creates and names a library, sets it as either STATIC or SHARED, and provides
# the relative paths to its source code. You can define multiple libraries, and
# CMake builds them for you. Gradle automatically packages shared libraries with
# your APK.
set(PaddleLite_DIR "${CMAKE_CURRENT_SOURCE_DIR}/../../../PaddleLite")
include_directories(${PaddleLite_DIR}/cxx/include)
set(OpenCV_DIR "${CMAKE_CURRENT_SOURCE_DIR}/../../../OpenCV/sdk/native/jni")
message(STATUS "opencv dir: ${OpenCV_DIR}")
find_package(OpenCV REQUIRED)
message(STATUS "OpenCV libraries: ${OpenCV_LIBS}")
include_directories(${OpenCV_INCLUDE_DIRS})
aux_source_directory(. SOURCES)
set(CMAKE_CXX_FLAGS
"${CMAKE_CXX_FLAGS} -ffast-math -Ofast -Os"
)
set(CMAKE_CXX_FLAGS
"${CMAKE_CXX_FLAGS} -fvisibility=hidden -fvisibility-inlines-hidden -fdata-sections -ffunction-sections"
)
set(CMAKE_SHARED_LINKER_FLAGS
"${CMAKE_SHARED_LINKER_FLAGS} -Wl,--gc-sections -Wl,-z,nocopyreloc")
add_library(
# Sets the name of the library.
Native
# Sets the library as a shared library.
SHARED
# Provides a relative path to your source file(s).
${SOURCES})
find_library(
# Sets the name of the path variable.
log-lib
# Specifies the name of the NDK library that you want CMake to locate.
log)
add_library(
# Sets the name of the library.
paddle_light_api_shared
# Sets the library as a shared library.
SHARED
# Provides a relative path to your source file(s).
IMPORTED)
set_target_properties(
# Specifies the target library.
paddle_light_api_shared
# Specifies the parameter you want to define.
PROPERTIES
IMPORTED_LOCATION
${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libpaddle_light_api_shared.so
# Provides the path to the library you want to import.
)
# Specifies libraries CMake should link to your target library. You can link
# multiple libraries, such as libraries you define in this build script,
# prebuilt third-party libraries, or system libraries.
target_link_libraries(
# Specifies the target library.
Native
paddle_light_api_shared
${OpenCV_LIBS}
GLESv2
EGL
jnigraphics
${log-lib}
)
add_custom_command(
TARGET Native
POST_BUILD
COMMAND
${CMAKE_COMMAND} -E copy
${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libc++_shared.so
${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libc++_shared.so)
add_custom_command(
TARGET Native
POST_BUILD
COMMAND
${CMAKE_COMMAND} -E copy
${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libpaddle_light_api_shared.so
${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libpaddle_light_api_shared.so)
add_custom_command(
TARGET Native
POST_BUILD
COMMAND
${CMAKE_COMMAND} -E copy
${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libhiai.so
${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libhiai.so)
add_custom_command(
TARGET Native
POST_BUILD
COMMAND
${CMAKE_COMMAND} -E copy
${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libhiai_ir.so
${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libhiai_ir.so)
add_custom_command(
TARGET Native
POST_BUILD
COMMAND
${CMAKE_COMMAND} -E copy
${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libhiai_ir_build.so
${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libhiai_ir_build.so)
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/common.h 0 → 100644
浏览文件 @ 26219d5f
//
// Created by fu on 4/25/18.
//
#pragma once
#import <vector>
#import <numeric>
#ifdef __ANDROID__
#include <android/log.h>
#define LOG_TAG "OCR_NDK"
#define LOGI(...) \
__android_log_print(ANDROID_LOG_INFO, LOG_TAG, __VA_ARGS__)
#define LOGW(...) \
__android_log_print(ANDROID_LOG_WARN, LOG_TAG, __VA_ARGS__)
#define LOGE(...) \
__android_log_print(ANDROID_LOG_ERROR, LOG_TAG, __VA_ARGS__)
#else
#include <stdio.h>
#define LOGI(format, ...) \
fprintf(stdout, "[" LOG_TAG "]" format "\n", ##__VA_ARGS__)
#define LOGW(format, ...) \
fprintf(stdout, "[" LOG_TAG "]" format "\n", ##__VA_ARGS__)
#define LOGE(format, ...) \
fprintf(stderr, "[" LOG_TAG "]Error: " format "\n", ##__VA_ARGS__)
#endif
enum RETURN_CODE {
RETURN_OK = 0
};
enum NET_TYPE{
NET_OCR = 900100,
NET_OCR_INTERNAL = 991008
};
template <typename T>
inline T product(const std::vector<T> &vec) {
if (vec.empty()){
return 0;
}
return std::accumulate(vec.begin(), vec.end(), 1, std::multiplies<T>());
}
deploy/android_demo/app/src/main/cpp/native.cpp 0 → 100644
浏览文件 @ 26219d5f
//
// Created by fujiayi on 2020/7/5.
//
#include "native.h"
#include "ocr_ppredictor.h"
#include <string>
#include <algorithm>
#include <paddle_api.h>
static paddle::lite_api::PowerMode str_to_cpu_mode(const std::string &cpu_mode);
extern "C"
JNIEXPORT jlong JNICALL
Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_init(JNIEnv *env, jobject thiz,
jstring j_det_model_path,
jstring j_rec_model_path,
jint j_thread_num,
jstring j_cpu_mode) {
std::string det_model_path = jstring_to_cpp_string(env, j_det_model_path);
std::string rec_model_path = jstring_to_cpp_string(env, j_rec_model_path);
int thread_num = j_thread_num;
std::string cpu_mode = jstring_to_cpp_string(env, j_cpu_mode);
ppredictor::OCR_Config conf;
conf.thread_num = thread_num;
conf.mode = str_to_cpu_mode(cpu_mode);
ppredictor::OCR_PPredictor *orc_predictor = new ppredictor::OCR_PPredictor{conf};
orc_predictor->init_from_file(det_model_path, rec_model_path);
return reinterpret_cast<jlong>(orc_predictor);
}
/**
* "LITE_POWER_HIGH" 转为 paddle::lite_api::LITE_POWER_HIGH
* @param cpu_mode
* @return
*/
static paddle::lite_api::PowerMode str_to_cpu_mode(const std::string &cpu_mode) {
static std::map<std::string, paddle::lite_api::PowerMode> cpu_mode_map{
{"LITE_POWER_HIGH", paddle::lite_api::LITE_POWER_HIGH},
{"LITE_POWER_LOW", paddle::lite_api::LITE_POWER_HIGH},
{"LITE_POWER_FULL", paddle::lite_api::LITE_POWER_FULL},
{"LITE_POWER_NO_BIND", paddle::lite_api::LITE_POWER_NO_BIND},
{"LITE_POWER_RAND_HIGH", paddle::lite_api::LITE_POWER_RAND_HIGH},
{"LITE_POWER_RAND_LOW", paddle::lite_api::LITE_POWER_RAND_LOW}
};
std::string upper_key;
std::transform(cpu_mode.cbegin(), cpu_mode.cend(), upper_key.begin(), ::toupper);
auto index = cpu_mode_map.find(upper_key);
if (index == cpu_mode_map.end()) {
LOGE("cpu_mode not found %s", upper_key.c_str());
return paddle::lite_api::LITE_POWER_HIGH;
} else {
return index->second;
}
}
extern "C"
JNIEXPORT jfloatArray JNICALL
Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_forward(JNIEnv *env, jobject thiz,
jlong java_pointer, jfloatArray buf,
jfloatArray ddims,
jobject original_image) {
LOGI("begin to run native forward");
if (java_pointer == 0) {
LOGE("JAVA pointer is NULL");
return cpp_array_to_jfloatarray(env, nullptr, 0);
}
cv::Mat origin = bitmap_to_cv_mat(env, original_image);
if (origin.size == 0) {
LOGE("origin bitmap cannot convert to CV Mat");
return cpp_array_to_jfloatarray(env, nullptr, 0);
}
ppredictor::OCR_PPredictor *ppredictor = (ppredictor::OCR_PPredictor *) java_pointer;
std::vector<float> dims_float_arr = jfloatarray_to_float_vector(env, ddims);
std::vector<int64_t> dims_arr;
dims_arr.resize(dims_float_arr.size());
std::copy(dims_float_arr.cbegin(), dims_float_arr.cend(), dims_arr.begin());
// 这里值有点大，就不调用jfloatarray_to_float_vector了
int64_t buf_len = (int64_t) env->GetArrayLength(buf);
jfloat *buf_data = env->GetFloatArrayElements(buf, JNI_FALSE);
float *data = (jfloat *) buf_data;
std::vector<ppredictor::OCRPredictResult> results = ppredictor->infer_ocr(dims_arr, data,
buf_len,
NET_OCR, origin);
LOGI("infer_ocr finished with boxes %ld", results.size());
// 这里将std::vector<ppredictor::OCRPredictResult> 序列化成 float数组，传输到java层再反序列化
std::vector<float> float_arr;
for (const ppredictor::OCRPredictResult &r :results) {
float_arr.push_back(r.points.size());
float_arr.push_back(r.word_index.size());
float_arr.push_back(r.score);
for (const std::vector<int> &point : r.points) {
float_arr.push_back(point.at(0));
float_arr.push_back(point.at(1));
}
for (int index: r.word_index) {
float_arr.push_back(index);
}
}
return cpp_array_to_jfloatarray(env, float_arr.data(), float_arr.size());
}
extern "C"
JNIEXPORT void JNICALL
Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_release(JNIEnv *env, jobject thiz,
jlong java_pointer){
if (java_pointer == 0) {
LOGE("JAVA pointer is NULL");
return;
}
ppredictor::OCR_PPredictor *ppredictor = (ppredictor::OCR_PPredictor *) java_pointer;
delete ppredictor;
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/native.h 0 → 100644
浏览文件 @ 26219d5f
//
// Created by fujiayi on 2020/7/5.
//
#pragma once
#include <jni.h>
#include <string>
#include <vector>
#include <android/bitmap.h>
#include <opencv2/opencv.hpp>
#include "common.h"
inline std::string jstring_to_cpp_string(JNIEnv *env, jstring jstr) {
// In java, a unicode char will be encoded using 2 bytes (utf16).
// so jstring will contain characters utf16. std::string in c++ is
// essentially a string of bytes, not characters, so if we want to
// pass jstring from JNI to c++, we have convert utf16 to bytes.
if (!jstr) {
return "";
}
const jclass stringClass = env->GetObjectClass(jstr);
const jmethodID getBytes =
env->GetMethodID(stringClass, "getBytes", "(Ljava/lang/String;)[B");
const jbyteArray stringJbytes = (jbyteArray) env->CallObjectMethod(
jstr, getBytes, env->NewStringUTF("UTF-8"));
size_t length = (size_t) env->GetArrayLength(stringJbytes);
jbyte *pBytes = env->GetByteArrayElements(stringJbytes, NULL);
std::string ret = std::string(reinterpret_cast<char *>(pBytes), length);
env->ReleaseByteArrayElements(stringJbytes, pBytes, JNI_ABORT);
env->DeleteLocalRef(stringJbytes);
env->DeleteLocalRef(stringClass);
return ret;
}
inline jstring cpp_string_to_jstring(JNIEnv *env, std::string str) {
auto *data = str.c_str();
jclass strClass = env->FindClass("java/lang/String");
jmethodID strClassInitMethodID =
env->GetMethodID(strClass, "<init>", "([BLjava/lang/String;)V");
jbyteArray bytes = env->NewByteArray(strlen(data));
env->SetByteArrayRegion(bytes, 0, strlen(data),
reinterpret_cast<const jbyte *>(data));
jstring encoding = env->NewStringUTF("UTF-8");
jstring res = (jstring) (
env->NewObject(strClass, strClassInitMethodID, bytes, encoding));
env->DeleteLocalRef(strClass);
env->DeleteLocalRef(encoding);
env->DeleteLocalRef(bytes);
return res;
}
inline jfloatArray cpp_array_to_jfloatarray(JNIEnv *env, const float *buf,
int64_t len) {
if (len == 0) {
return env->NewFloatArray(0);
}
jfloatArray result = env->NewFloatArray(len);
env->SetFloatArrayRegion(result, 0, len, buf);
return result;
}
inline jintArray cpp_array_to_jintarray(JNIEnv *env, const int *buf,
int64_t len) {
jintArray result = env->NewIntArray(len);
env->SetIntArrayRegion(result, 0, len, buf);
return result;
}
inline jbyteArray cpp_array_to_jbytearray(JNIEnv *env, const int8_t *buf,
int64_t len) {
jbyteArray result = env->NewByteArray(len);
env->SetByteArrayRegion(result, 0, len, buf);
return result;
}
inline jlongArray int64_vector_to_jlongarray(JNIEnv *env,
const std::vector<int64_t> &vec) {
jlongArray result = env->NewLongArray(vec.size());
jlong *buf = new jlong[vec.size()];
for (size_t i = 0; i < vec.size(); ++i) {
buf[i] = (jlong) vec[i];
}
env->SetLongArrayRegion(result, 0, vec.size(), buf);
delete[] buf;
return result;
}
inline std::vector<int64_t> jlongarray_to_int64_vector(JNIEnv *env,
jlongArray data) {
int data_size = env->GetArrayLength(data);
jlong *data_ptr = env->GetLongArrayElements(data, nullptr);
std::vector<int64_t> data_vec(data_ptr, data_ptr + data_size);
env->ReleaseLongArrayElements(data, data_ptr, 0);
return data_vec;
}
inline std::vector<float> jfloatarray_to_float_vector(JNIEnv *env,
jfloatArray data) {
int data_size = env->GetArrayLength(data);
jfloat *data_ptr = env->GetFloatArrayElements(data, nullptr);
std::vector<float> data_vec(data_ptr, data_ptr + data_size);
env->ReleaseFloatArrayElements(data, data_ptr, 0);
return data_vec;
}
inline cv::Mat bitmap_to_cv_mat(JNIEnv *env, jobject bitmap) {
AndroidBitmapInfo info;
int result = AndroidBitmap_getInfo(env, bitmap, &info);
if (result != ANDROID_BITMAP_RESULT_SUCCESS) {
LOGE("AndroidBitmap_getInfo failed, result: %d", result);
return cv::Mat{};
}
if (info.format != ANDROID_BITMAP_FORMAT_RGBA_8888) {
LOGE("Bitmap format is not RGBA_8888 !");
return cv::Mat{};
}
unsigned char *srcData = NULL;
AndroidBitmap_lockPixels(env, bitmap, (void **) &srcData);
cv::Mat mat = cv::Mat::zeros(info.height, info.width, CV_8UC4);
memcpy(mat.data, srcData, info.height * info.width * 4);
AndroidBitmap_unlockPixels(env, bitmap);
cv::cvtColor(mat, mat, cv::COLOR_RGBA2BGR);
/**
if (!cv::imwrite("/sdcard/1/copy.jpg", mat)){
LOGE("Write image failed " );
}
*/
return mat;
}
deploy/android_demo/app/src/main/cpp/ocr_clipper.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/android_demo/app/src/main/cpp/ocr_clipper.hpp 0 → 100644
浏览文件 @ 26219d5f
/*******************************************************************************
* *
* Author : Angus Johnson *
* Version : 6.4.2 *
* Date : 27 February 2017 *
* Website : http://www.angusj.com *
* Copyright : Angus Johnson 2010-2017 *
* *
* License: *
* Use, modification & distribution is subject to Boost Software License Ver 1. *
* http://www.boost.org/LICENSE_1_0.txt *
* *
* Attributions: *
* The code in this library is an extension of Bala Vatti's clipping algorithm: *
* "A generic solution to polygon clipping" *
* Communications of the ACM, Vol 35, Issue 7 (July 1992) pp 56-63. *
* http://portal.acm.org/citation.cfm?id=129906 *
* *
* Computer graphics and geometric modeling: implementation and algorithms *
* By Max K. Agoston *
* Springer; 1 edition (January 4, 2005) *
* http://books.google.com/books?q=vatti+clipping+agoston *
* *
* See also: *
* "Polygon Offsetting by Computing Winding Numbers" *
* Paper no. DETC2005-85513 pp. 565-575 *
* ASME 2005 International Design Engineering Technical Conferences *
* and Computers and Information in Engineering Conference (IDETC/CIE2005) *
* September 24-28, 2005 , Long Beach, California, USA *
* http://www.me.berkeley.edu/~mcmains/pubs/DAC05OffsetPolygon.pdf *
* *
*******************************************************************************/
#ifndef clipper_hpp
#define clipper_hpp
#define CLIPPER_VERSION "6.4.2"
//use_int32: When enabled 32bit ints are used instead of 64bit ints. This
//improve performance but coordinate values are limited to the range +/- 46340
//#define use_int32
//use_xyz: adds a Z member to IntPoint. Adds a minor cost to perfomance.
//#define use_xyz
//use_lines: Enables line clipping. Adds a very minor cost to performance.
#define use_lines
//use_deprecated: Enables temporary support for the obsolete functions
//#define use_deprecated
#include <vector>
#include <list>
#include <set>
#include <stdexcept>
#include <cstring>
#include <cstdlib>
#include <ostream>
#include <functional>
#include <queue>
namespace ClipperLib {
enum ClipType {
ctIntersection, ctUnion, ctDifference, ctXor
};
enum PolyType {
ptSubject, ptClip
};
//By far the most widely used winding rules for polygon filling are
//EvenOdd & NonZero (GDI, GDI+, XLib, OpenGL, Cairo, AGG, Quartz, SVG, Gr32)
//Others rules include Positive, Negative and ABS_GTR_EQ_TWO (only in OpenGL)
//see http://glprogramming.com/red/chapter11.html
enum PolyFillType {
pftEvenOdd, pftNonZero, pftPositive, pftNegative
};
#ifdef use_int32
typedef int cInt;
static cInt const loRange = 0x7FFF;
static cInt const hiRange = 0x7FFF;
#else
typedef signed long long cInt;
static cInt const loRange = 0x3FFFFFFF;
static cInt const hiRange = 0x3FFFFFFFFFFFFFFFLL;
typedef signed long long long64; //used by Int128 class
typedef unsigned long long ulong64;
#endif
struct IntPoint {
cInt X;
cInt Y;
#ifdef use_xyz
cInt Z;
IntPoint(cInt x = 0, cInt y = 0, cInt z = 0): X(x), Y(y), Z(z) {};
#else
IntPoint(cInt x = 0, cInt y = 0) : X(x), Y(y) {};
#endif
friend inline bool operator==(const IntPoint &a, const IntPoint &b) {
return a.X == b.X && a.Y == b.Y;
}
friend inline bool operator!=(const IntPoint &a, const IntPoint &b) {
return a.X != b.X || a.Y != b.Y;
}
};
//------------------------------------------------------------------------------
typedef std::vector <IntPoint> Path;
typedef std::vector <Path> Paths;
inline Path &operator<<(Path &poly, const IntPoint &p) {
poly.push_back(p);
return poly;
}
inline Paths &operator<<(Paths &polys, const Path &p) {
polys.push_back(p);
return polys;
}
std::ostream &operator<<(std::ostream &s, const IntPoint &p);
std::ostream &operator<<(std::ostream &s, const Path &p);
std::ostream &operator<<(std::ostream &s, const Paths &p);
struct DoublePoint {
double X;
double Y;
DoublePoint(double x = 0, double y = 0) : X(x), Y(y) {}
DoublePoint(IntPoint ip) : X((double) ip.X), Y((double) ip.Y) {}
};
//------------------------------------------------------------------------------
#ifdef use_xyz
typedef void (*ZFillCallback)(IntPoint& e1bot, IntPoint& e1top, IntPoint& e2bot, IntPoint& e2top, IntPoint& pt);
#endif
enum InitOptions {
ioReverseSolution = 1, ioStrictlySimple = 2, ioPreserveCollinear = 4
};
enum JoinType {
jtSquare, jtRound, jtMiter
};
enum EndType {
etClosedPolygon, etClosedLine, etOpenButt, etOpenSquare, etOpenRound
};
class PolyNode;
typedef std::vector<PolyNode *> PolyNodes;
class PolyNode {
public:
PolyNode();
virtual ~PolyNode() {};
Path Contour;
PolyNodes Childs;
PolyNode *Parent;
PolyNode *GetNext() const;
bool IsHole() const;
bool IsOpen() const;
int ChildCount() const;
private:
//PolyNode& operator =(PolyNode& other);
unsigned Index; //node index in Parent.Childs
bool m_IsOpen;
JoinType m_jointype;
EndType m_endtype;
PolyNode *GetNextSiblingUp() const;
void AddChild(PolyNode &child);
friend class Clipper; //to access Index
friend class ClipperOffset;
};
class PolyTree : public PolyNode {
public:
~PolyTree() { Clear(); };
PolyNode *GetFirst() const;
void Clear();
int Total() const;
private:
//PolyTree& operator =(PolyTree& other);
PolyNodes AllNodes;
friend class Clipper; //to access AllNodes
};
bool Orientation(const Path &poly);
double Area(const Path &poly);
int PointInPolygon(const IntPoint &pt, const Path &path);
void SimplifyPolygon(const Path &in_poly, Paths &out_polys, PolyFillType fillType = pftEvenOdd);
void SimplifyPolygons(const Paths &in_polys, Paths &out_polys, PolyFillType fillType = pftEvenOdd);
void SimplifyPolygons(Paths &polys, PolyFillType fillType = pftEvenOdd);
void CleanPolygon(const Path &in_poly, Path &out_poly, double distance = 1.415);
void CleanPolygon(Path &poly, double distance = 1.415);
void CleanPolygons(const Paths &in_polys, Paths &out_polys, double distance = 1.415);
void CleanPolygons(Paths &polys, double distance = 1.415);
void MinkowskiSum(const Path &pattern, const Path &path, Paths &solution, bool pathIsClosed);
void MinkowskiSum(const Path &pattern, const Paths &paths, Paths &solution, bool pathIsClosed);
void MinkowskiDiff(const Path &poly1, const Path &poly2, Paths &solution);
void PolyTreeToPaths(const PolyTree &polytree, Paths &paths);
void ClosedPathsFromPolyTree(const PolyTree &polytree, Paths &paths);
void OpenPathsFromPolyTree(PolyTree &polytree, Paths &paths);
void ReversePath(Path &p);
void ReversePaths(Paths &p);
struct IntRect {
cInt left;
cInt top;
cInt right;
cInt bottom;
};
//enums that are used internally ...
enum EdgeSide {
esLeft = 1, esRight = 2
};
//forward declarations (for stuff used internally) ...
struct TEdge;
struct IntersectNode;
struct LocalMinimum;
struct OutPt;
struct OutRec;
struct Join;
typedef std::vector<OutRec *> PolyOutList;
typedef std::vector<TEdge *> EdgeList;
typedef std::vector<Join *> JoinList;
typedef std::vector<IntersectNode *> IntersectList;
//------------------------------------------------------------------------------
//ClipperBase is the ancestor to the Clipper class. It should not be
//instantiated directly. This class simply abstracts the conversion of sets of
//polygon coordinates into edge objects that are stored in a LocalMinima list.
class ClipperBase {
public:
ClipperBase();
virtual ~ClipperBase();
virtual bool AddPath(const Path &pg, PolyType PolyTyp, bool Closed);
bool AddPaths(const Paths &ppg, PolyType PolyTyp, bool Closed);
virtual void Clear();
IntRect GetBounds();
bool PreserveCollinear() { return m_PreserveCollinear; };
void PreserveCollinear(bool value) { m_PreserveCollinear = value; };
protected:
void DisposeLocalMinimaList();
TEdge *AddBoundsToLML(TEdge *e, bool IsClosed);
virtual void Reset();
TEdge *ProcessBound(TEdge *E, bool IsClockwise);
void InsertScanbeam(const cInt Y);
bool PopScanbeam(cInt &Y);
bool LocalMinimaPending();
bool PopLocalMinima(cInt Y, const LocalMinimum *&locMin);
OutRec *CreateOutRec();
void DisposeAllOutRecs();
void DisposeOutRec(PolyOutList::size_type index);
void SwapPositionsInAEL(TEdge *edge1, TEdge *edge2);
void DeleteFromAEL(TEdge *e);
void UpdateEdgeIntoAEL(TEdge *&e);
typedef std::vector <LocalMinimum> MinimaList;
MinimaList::iterator m_CurrentLM;
MinimaList m_MinimaList;
bool m_UseFullRange;
EdgeList m_edges;
bool m_PreserveCollinear;
bool m_HasOpenPaths;
PolyOutList m_PolyOuts;
TEdge *m_ActiveEdges;
typedef std::priority_queue <cInt> ScanbeamList;
ScanbeamList m_Scanbeam;
};
//------------------------------------------------------------------------------
class Clipper : public virtual ClipperBase {
public:
Clipper(int initOptions = 0);
bool Execute(ClipType clipType,
Paths &solution,
PolyFillType fillType = pftEvenOdd);
bool Execute(ClipType clipType,
Paths &solution,
PolyFillType subjFillType,
PolyFillType clipFillType);
bool Execute(ClipType clipType,
PolyTree &polytree,
PolyFillType fillType = pftEvenOdd);
bool Execute(ClipType clipType,
PolyTree &polytree,
PolyFillType subjFillType,
PolyFillType clipFillType);
bool ReverseSolution() { return m_ReverseOutput; };
void ReverseSolution(bool value) { m_ReverseOutput = value; };
bool StrictlySimple() { return m_StrictSimple; };
void StrictlySimple(bool value) { m_StrictSimple = value; };
//set the callback function for z value filling on intersections (otherwise Z is 0)
#ifdef use_xyz
void ZFillFunction(ZFillCallback zFillFunc);
#endif
protected:
virtual bool ExecuteInternal();
private:
JoinList m_Joins;
JoinList m_GhostJoins;
IntersectList m_IntersectList;
ClipType m_ClipType;
typedef std::list <cInt> MaximaList;
MaximaList m_Maxima;
TEdge *m_SortedEdges;
bool m_ExecuteLocked;
PolyFillType m_ClipFillType;
PolyFillType m_SubjFillType;
bool m_ReverseOutput;
bool m_UsingPolyTree;
bool m_StrictSimple;
#ifdef use_xyz
ZFillCallback m_ZFill; //custom callback
#endif
void SetWindingCount(TEdge &edge);
bool IsEvenOddFillType(const TEdge &edge) const;
bool IsEvenOddAltFillType(const TEdge &edge) const;
void InsertLocalMinimaIntoAEL(const cInt botY);
void InsertEdgeIntoAEL(TEdge *edge, TEdge *startEdge);
void AddEdgeToSEL(TEdge *edge);
bool PopEdgeFromSEL(TEdge *&edge);
void CopyAELToSEL();
void DeleteFromSEL(TEdge *e);
void SwapPositionsInSEL(TEdge *edge1, TEdge *edge2);
bool IsContributing(const TEdge &edge) const;
bool IsTopHorz(const cInt XPos);
void DoMaxima(TEdge *e);
void ProcessHorizontals();
void ProcessHorizontal(TEdge *horzEdge);
void AddLocalMaxPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
OutPt *AddLocalMinPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
OutRec *GetOutRec(int idx);
void AppendPolygon(TEdge *e1, TEdge *e2);
void IntersectEdges(TEdge *e1, TEdge *e2, IntPoint &pt);
OutPt *AddOutPt(TEdge *e, const IntPoint &pt);
OutPt *GetLastOutPt(TEdge *e);
bool ProcessIntersections(const cInt topY);
void BuildIntersectList(const cInt topY);
void ProcessIntersectList();
void ProcessEdgesAtTopOfScanbeam(const cInt topY);
void BuildResult(Paths &polys);
void BuildResult2(PolyTree &polytree);
void SetHoleState(TEdge *e, OutRec *outrec);
void DisposeIntersectNodes();
bool FixupIntersectionOrder();
void FixupOutPolygon(OutRec &outrec);
void FixupOutPolyline(OutRec &outrec);
bool IsHole(TEdge *e);
bool FindOwnerFromSplitRecs(OutRec &outRec, OutRec *&currOrfl);
void FixHoleLinkage(OutRec &outrec);
void AddJoin(OutPt *op1, OutPt *op2, const IntPoint offPt);
void ClearJoins();
void ClearGhostJoins();
void AddGhostJoin(OutPt *op, const IntPoint offPt);
bool JoinPoints(Join *j, OutRec *outRec1, OutRec *outRec2);
void JoinCommonEdges();
void DoSimplePolygons();
void FixupFirstLefts1(OutRec *OldOutRec, OutRec *NewOutRec);
void FixupFirstLefts2(OutRec *InnerOutRec, OutRec *OuterOutRec);
void FixupFirstLefts3(OutRec *OldOutRec, OutRec *NewOutRec);
#ifdef use_xyz
void SetZ(IntPoint& pt, TEdge& e1, TEdge& e2);
#endif
};
//------------------------------------------------------------------------------
class ClipperOffset {
public:
ClipperOffset(double miterLimit = 2.0, double roundPrecision = 0.25);
~ClipperOffset();
void AddPath(const Path &path, JoinType joinType, EndType endType);
void AddPaths(const Paths &paths, JoinType joinType, EndType endType);
void Execute(Paths &solution, double delta);
void Execute(PolyTree &solution, double delta);
void Clear();
double MiterLimit;
double ArcTolerance;
private:
Paths m_destPolys;
Path m_srcPoly;
Path m_destPoly;
std::vector <DoublePoint> m_normals;
double m_delta, m_sinA, m_sin, m_cos;
double m_miterLim, m_StepsPerRad;
IntPoint m_lowest;
PolyNode m_polyNodes;
void FixOrientations();
void DoOffset(double delta);
void OffsetPoint(int j, int &k, JoinType jointype);
void DoSquare(int j, int k);
void DoMiter(int j, int k, double r);
void DoRound(int j, int k);
};
//------------------------------------------------------------------------------
class clipperException : public std::exception {
public:
clipperException(const char *description) : m_descr(description) {}
virtual ~clipperException() throw() {}
virtual const char *what() const throw() { return m_descr.c_str(); }
private:
std::string m_descr;
};
//------------------------------------------------------------------------------
} //ClipperLib namespace
#endif //clipper_hpp
deploy/android_demo/app/src/main/cpp/ocr_crnn_process.cpp 0 → 100644
浏览文件 @ 26219d5f
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "ocr_crnn_process.h"
#include <iostream>
#include <vector>
#include <iostream>
#include <cstring>
#include <fstream>
#include <cmath>
const std::string CHARACTER_TYPE = "ch";
const int MAX_DICT_LENGTH = 6624;
const std::vector<int> REC_IMAGE_SHAPE = {3, 32, 320};
static cv::Mat crnn_resize_norm_img(cv::Mat img, float wh_ratio) {
int imgC = REC_IMAGE_SHAPE[0];
int imgW = REC_IMAGE_SHAPE[2];
int imgH = REC_IMAGE_SHAPE[1];
if (CHARACTER_TYPE == "ch")
imgW = int(32 * wh_ratio);
float ratio = float(img.cols) / float(img.rows);
int resize_w = 0;
if (ceilf(imgH * ratio) > imgW)
resize_w = imgW;
else
resize_w = int(ceilf(imgH * ratio));
cv::Mat resize_img;
cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f, cv::INTER_CUBIC);
resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
for (int h = 0; h < resize_img.rows; h++) {
for (int w = 0; w < resize_img.cols; w++) {
resize_img.at<cv::Vec3f>(h, w)[0] = (resize_img.at<cv::Vec3f>(h, w)[0] - 0.5) * 2;
resize_img.at<cv::Vec3f>(h, w)[1] = (resize_img.at<cv::Vec3f>(h, w)[1] - 0.5) * 2;
resize_img.at<cv::Vec3f>(h, w)[2] = (resize_img.at<cv::Vec3f>(h, w)[2] - 0.5) * 2;
}
}
cv::Mat dist;
cv::copyMakeBorder(resize_img, dist, 0, 0, 0, int(imgW - resize_w), cv::BORDER_CONSTANT,
{0, 0, 0});
return dist;
}
cv::Mat crnn_resize_img(const cv::Mat &img, float wh_ratio) {
int imgC = REC_IMAGE_SHAPE[0];
int imgW = REC_IMAGE_SHAPE[2];
int imgH = REC_IMAGE_SHAPE[1];
if (CHARACTER_TYPE == "ch") {
imgW = int(32 * wh_ratio);
}
float ratio = float(img.cols) / float(img.rows);
int resize_w = 0;
if (ceilf(imgH * ratio) > imgW)
resize_w = imgW;
else
resize_w = int(ceilf(imgH * ratio));
cv::Mat resize_img;
cv::resize(img, resize_img, cv::Size(resize_w, imgH));
return resize_img;
}
cv::Mat get_rotate_crop_image(const cv::Mat &srcimage, const std::vector<std::vector<int>> &box) {
std::vector<std::vector<int>> points = box;
int x_collect[4] = {box[0][0], box[1][0], box[2][0], box[3][0]};
int y_collect[4] = {box[0][1], box[1][1], box[2][1], box[3][1]};
int left = int(*std::min_element(x_collect, x_collect + 4));
int right = int(*std::max_element(x_collect, x_collect + 4));
int top = int(*std::min_element(y_collect, y_collect + 4));
int bottom = int(*std::max_element(y_collect, y_collect + 4));
cv::Mat img_crop;
srcimage(cv::Rect(left, top, right - left, bottom - top)).copyTo(img_crop);
for (int i = 0; i < points.size(); i++) {
points[i][0] -= left;
points[i][1] -= top;
}
int img_crop_width = int(sqrt(pow(points[0][0] - points[1][0], 2) +
pow(points[0][1] - points[1][1], 2)));
int img_crop_height = int(sqrt(pow(points[0][0] - points[3][0], 2) +
pow(points[0][1] - points[3][1], 2)));
cv::Point2f pts_std[4];
pts_std[0] = cv::Point2f(0., 0.);
pts_std[1] = cv::Point2f(img_crop_width, 0.);
pts_std[2] = cv::Point2f(img_crop_width, img_crop_height);
pts_std[3] = cv::Point2f(0.f, img_crop_height);
cv::Point2f pointsf[4];
pointsf[0] = cv::Point2f(points[0][0], points[0][1]);
pointsf[1] = cv::Point2f(points[1][0], points[1][1]);
pointsf[2] = cv::Point2f(points[2][0], points[2][1]);
pointsf[3] = cv::Point2f(points[3][0], points[3][1]);
cv::Mat M = cv::getPerspectiveTransform(pointsf, pts_std);
cv::Mat dst_img;
cv::warpPerspective(img_crop, dst_img, M, cv::Size(img_crop_width, img_crop_height),
cv::BORDER_REPLICATE);
if (float(dst_img.rows) >= float(dst_img.cols) * 1.5) {
/*
cv::Mat srcCopy = cv::Mat(dst_img.rows, dst_img.cols, dst_img.depth());
cv::transpose(dst_img, srcCopy);
cv::flip(srcCopy, srcCopy, 0);
return srcCopy;
*/
cv::transpose(dst_img, dst_img);
cv::flip(dst_img, dst_img, 0);
return dst_img;
} else {
return dst_img;
}
}
deploy/android_demo/app/src/main/cpp/ocr_crnn_process.h 0 → 100644
浏览文件 @ 26219d5f
//
// Created by fujiayi on 2020/7/3.
//
#pragma once
#include <vector>
#include <opencv2/opencv.hpp>
#include "common.h"
extern const std::vector<int> REC_IMAGE_SHAPE;
cv::Mat get_rotate_crop_image(const cv::Mat &srcimage, const std::vector<std::vector<int>> &box);
cv::Mat crnn_resize_img(const cv::Mat &img, float wh_ratio);
template<class ForwardIterator>
inline size_t argmax(ForwardIterator first, ForwardIterator last) {
return std::distance(first, std::max_element(first, last));
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/ocr_db_post_process.cpp 0 → 100644
浏览文件 @ 26219d5f
// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include <iostream>
#include <vector>
#include <math.h>
#include "opencv2/core.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/imgproc.hpp"
#include "ocr_clipper.hpp"
static void getcontourarea(float **box, float unclip_ratio, float &distance) {
int pts_num = 4;
float area = 0.0f;
float dist = 0.0f;
for (int i = 0; i < pts_num; i++) {
area += box[i][0] * box[(i + 1) % pts_num][1] - box[i][1] * box[(i + 1) % pts_num][0];
dist += sqrtf(
(box[i][0] - box[(i + 1) % pts_num][0]) * (box[i][0] - box[(i + 1) % pts_num][0]) +
(box[i][1] - box[(i + 1) % pts_num][1]) * (box[i][1] - box[(i + 1) % pts_num][1]));
}
area = fabs(float(area / 2.0));
distance = area * unclip_ratio / dist;
}
static cv::RotatedRect unclip(float **box) {
float unclip_ratio = 2.0;
float distance = 1.0;
getcontourarea(box, unclip_ratio, distance);
ClipperLib::ClipperOffset offset;
ClipperLib::Path p;
p << ClipperLib::IntPoint(int(box[0][0]), int(box[0][1]))
<< ClipperLib::IntPoint(int(box[1][0]), int(box[1][1])) <<
ClipperLib::IntPoint(int(box[2][0]), int(box[2][1]))
<< ClipperLib::IntPoint(int(box[3][0]), int(box[3][1]));
offset.AddPath(p, ClipperLib::jtRound, ClipperLib::etClosedPolygon);
ClipperLib::Paths soln;
offset.Execute(soln, distance);
std::vector<cv::Point2f> points;
for (int j = 0; j < soln.size(); j++) {
for (int i = 0; i < soln[soln.size() - 1].size(); i++) {
points.emplace_back(soln[j][i].X, soln[j][i].Y);
}
}
cv::RotatedRect res = cv::minAreaRect(points);
return res;
}
static float **Mat2Vec(cv::Mat mat) {
auto **array = new float *[mat.rows];
for (int i = 0; i < mat.rows; ++i){
array[i] = new float[mat.cols];
}
for (int i = 0; i < mat.rows; ++i) {
for (int j = 0; j < mat.cols; ++j) {
array[i][j] = mat.at<float>(i, j);
}
}
return array;
}
static void quickSort(float **s, int l, int r) {
if (l < r) {
int i = l, j = r;
float x = s[l][0];
float *xp = s[l];
while (i < j) {
while (i < j && s[j][0] >= x){
j--;
}
if (i < j){
std::swap(s[i++], s[j]);
}
while (i < j && s[i][0] < x){
i++;
}
if (i < j){
std::swap(s[j--], s[i]);
}
}
s[i] = xp;
quickSort(s, l, i - 1);
quickSort(s, i + 1, r);
}
}
static void quickSort_vector(std::vector<std::vector<int>> &box, int l, int r, int axis) {
if (l < r) {
int i = l, j = r;
int x = box[l][axis];
std::vector<int> xp(box[l]);
while (i < j) {
while (i < j && box[j][axis] >= x){
j--;
}
if (i < j){
std::swap(box[i++], box[j]);
}
while (i < j && box[i][axis] < x){
i++;
}
if (i < j){
std::swap(box[j--], box[i]);
}
}
box[i] = xp;
quickSort_vector(box, l, i - 1, axis);
quickSort_vector(box, i + 1, r, axis);
}
}
static std::vector<std::vector<int>> order_points_clockwise(std::vector<std::vector<int>> pts) {
std::vector<std::vector<int>> box = pts;
quickSort_vector(box, 0, int(box.size() - 1), 0);
std::vector<std::vector<int>> leftmost = {box[0], box[1]};
std::vector<std::vector<int>> rightmost = {box[2], box[3]};
if (leftmost[0][1] > leftmost[1][1]){
std::swap(leftmost[0], leftmost[1]);
}
if (rightmost[0][1] > rightmost[1][1]){
std::swap(rightmost[0], rightmost[1]);
}
std::vector<std::vector<int>> rect = {leftmost[0], rightmost[0], rightmost[1], leftmost[1]};
return rect;
}
static float **get_mini_boxes(cv::RotatedRect box, float &ssid) {
ssid = box.size.width >= box.size.height ? box.size.height : box.size.width;
cv::Mat points;
cv::boxPoints(box, points);
// sorted box points
auto array = Mat2Vec(points);
quickSort(array, 0, 3);
float *idx1 = array[0], *idx2 = array[1], *idx3 = array[2], *idx4 = array[3];
if (array[3][1] <= array[2][1]) {
idx2 = array[3];
idx3 = array[2];
} else {
idx2 = array[2];
idx3 = array[3];
}
if (array[1][1] <= array[0][1]) {
idx1 = array[1];
idx4 = array[0];
} else {
idx1 = array[0];
idx4 = array[1];
}
array[0] = idx1;
array[1] = idx2;
array[2] = idx3;
array[3] = idx4;
return array;
}
template <class T> T clamp(T x, T min, T max) {
if (x > max){
return max;
}
if (x < min){
return min;
}
return x;
}
static float clampf(float x, float min, float max) {
if (x > max)
return max;
if (x < min)
return min;
return x;
}
float box_score_fast(float **box_array, cv::Mat pred) {
auto array = box_array;
int width = pred.cols;
int height = pred.rows;
float box_x[4] = {array[0][0], array[1][0], array[2][0], array[3][0]};
float box_y[4] = {array[0][1], array[1][1], array[2][1], array[3][1]};
int xmin = clamp(int(std::floorf(*(std::min_element(box_x, box_x + 4)))), 0, width - 1);
int xmax = clamp(int(std::ceilf(*(std::max_element(box_x, box_x + 4)))), 0, width - 1);
int ymin = clamp(int(std::floorf(*(std::min_element(box_y, box_y + 4)))), 0, height - 1);
int ymax = clamp(int(std::ceilf(*(std::max_element(box_y, box_y + 4)))), 0, height - 1);
cv::Mat mask;
mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
cv::Point root_point[4];
root_point[0] = cv::Point(int(array[0][0]) - xmin, int(array[0][1]) - ymin);
root_point[1] = cv::Point(int(array[1][0]) - xmin, int(array[1][1]) - ymin);
root_point[2] = cv::Point(int(array[2][0]) - xmin, int(array[2][1]) - ymin);
root_point[3] = cv::Point(int(array[3][0]) - xmin, int(array[3][1]) - ymin);
const cv::Point *ppt[1] = {root_point};
int npt[] = {4};
cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
cv::Mat croppedImg;
pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1)).copyTo(croppedImg);
auto score = cv::mean(croppedImg, mask)[0];
return score;
}
std::vector<std::vector<std::vector<int>>>
boxes_from_bitmap(const cv::Mat& pred, const cv::Mat& bitmap) {
const int min_size = 3;
const int max_candidates = 1000;
const float box_thresh = 0.5;
int width = bitmap.cols;
int height = bitmap.rows;
std::vector<std::vector<cv::Point>> contours;
std::vector<cv::Vec4i> hierarchy;
cv::findContours(bitmap, contours, hierarchy, cv::RETR_LIST, cv::CHAIN_APPROX_SIMPLE);
int num_contours = contours.size() >= max_candidates ? max_candidates : contours.size();
std::vector<std::vector<std::vector<int>>> boxes;
for (int _i = 0; _i < num_contours; _i++) {
float ssid;
cv::RotatedRect box = cv::minAreaRect(contours[_i]);
auto array = get_mini_boxes(box, ssid);
auto box_for_unclip = array;
//end get_mini_box
if (ssid < min_size) {
continue;
}
float score;
score = box_score_fast(array, pred);
//end box_score_fast
if (score < box_thresh){
continue;
}
// start for unclip
cv::RotatedRect points = unclip(box_for_unclip);
// end for unclip
cv::RotatedRect clipbox = points;
auto cliparray = get_mini_boxes(clipbox, ssid);
if (ssid < min_size + 2) continue;
int dest_width = pred.cols;
int dest_height = pred.rows;
std::vector<std::vector<int>> intcliparray;
for (int num_pt = 0; num_pt < 4; num_pt++) {
std::vector<int> a{
int(clampf(roundf(cliparray[num_pt][0] / float(width) * float(dest_width)), 0,
float(dest_width))),
int(clampf(roundf(cliparray[num_pt][1] / float(height) * float(dest_height)), 0,
float(dest_height)))};
intcliparray.emplace_back(std::move(a));
}
boxes.emplace_back(std::move(intcliparray));
}//end for
return boxes;
}
int _max(int a, int b) {
return a >= b ? a : b;
}
int _min(int a, int b) {
return a >= b ? b : a;
}
std::vector<std::vector<std::vector<int>>>
filter_tag_det_res(const std::vector<std::vector<std::vector<int>>>& o_boxes,
float ratio_h, float ratio_w,const cv::Mat& srcimg) {
int oriimg_h = srcimg.rows;
int oriimg_w = srcimg.cols;
std::vector<std::vector<std::vector<int>>> boxes{o_boxes};
std::vector<std::vector<std::vector<int>>> root_points;
for (int n = 0; n < boxes.size(); n++) {
boxes[n] = order_points_clockwise(boxes[n]);
for (int m = 0; m < boxes[0].size(); m++) {
boxes[n][m][0] /= ratio_w;
boxes[n][m][1] /= ratio_h;
boxes[n][m][0] = int(_min(_max(boxes[n][m][0], 0), oriimg_w - 1));
boxes[n][m][1] = int(_min(_max(boxes[n][m][1], 0), oriimg_h - 1));
}
}
for (int n = 0; n < boxes.size(); n++) {
int rect_width, rect_height;
rect_width = int(sqrt(
pow(boxes[n][0][0] - boxes[n][1][0], 2) + pow(boxes[n][0][1] - boxes[n][1][1], 2)));
rect_height = int(sqrt(
pow(boxes[n][0][0] - boxes[n][3][0], 2) + pow(boxes[n][0][1] - boxes[n][3][1], 2)));
if (rect_width <= 10 || rect_height <= 10)
continue;
root_points.push_back(boxes[n]);
}
return root_points;
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/ocr_db_post_process.h 0 → 100644
浏览文件 @ 26219d5f
//
// Created by fujiayi on 2020/7/2.
//
#pragma once
#include <vector>
#include <opencv2/opencv.hpp>
std::vector<std::vector<std::vector<int>>>
boxes_from_bitmap(const cv::Mat &pred, const cv::Mat &bitmap);
std::vector<std::vector<std::vector<int>>>
filter_tag_det_res(
const std::vector<std::vector<std::vector<int>>> &o_boxes,
float ratio_h,
float ratio_w,
const cv::Mat &srcimg
);
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/ocr_ppredictor.cpp 0 → 100644
浏览文件 @ 26219d5f
//
// Created by fujiayi on 2020/7/1.
//
#include "ocr_ppredictor.h"
#include "preprocess.h"
#include "common.h"
#include "ocr_db_post_process.h"
#include "ocr_crnn_process.h"
namespace ppredictor {
OCR_PPredictor::OCR_PPredictor(const OCR_Config &config) : _config(config) {
}
int
OCR_PPredictor::init(const std::string &det_model_content, const std::string &rec_model_content) {
_det_predictor = std::unique_ptr<PPredictor>(
new PPredictor{_config.thread_num, NET_OCR, _config.mode});
_det_predictor->init_nb(det_model_content);
_rec_predictor = std::unique_ptr<PPredictor>(
new PPredictor{_config.thread_num, NET_OCR_INTERNAL, _config.mode});
_rec_predictor->init_nb(rec_model_content);
return RETURN_OK;
}
int OCR_PPredictor::init_from_file(const std::string &det_model_path, const std::string &rec_model_path){
_det_predictor = std::unique_ptr<PPredictor>(
new PPredictor{_config.thread_num, NET_OCR, _config.mode});
_det_predictor->init_from_file(det_model_path);
_rec_predictor = std::unique_ptr<PPredictor>(
new PPredictor{_config.thread_num, NET_OCR_INTERNAL, _config.mode});
_rec_predictor->init_from_file(rec_model_path);
return RETURN_OK;
}
/**
* 调试用，保存第一步的框选结果
* @param filter_boxes
* @param boxes
* @param srcimg
*/
static void visual_img(const std::vector<std::vector<std::vector<int>>> &filter_boxes,
const std::vector<std::vector<std::vector<int>>> &boxes,
const cv::Mat &srcimg) {
// visualization
cv::Point rook_points[filter_boxes.size()][4];
for (int n = 0; n < filter_boxes.size(); n++) {
for (int m = 0; m < filter_boxes[0].size(); m++) {
rook_points[n][m] = cv::Point(int(filter_boxes[n][m][0]), int(filter_boxes[n][m][1]));
}
}
cv::Mat img_vis;
srcimg.copyTo(img_vis);
for (int n = 0; n < boxes.size(); n++) {
const cv::Point *ppt[1] = {rook_points[n]};
int npt[] = {4};
cv::polylines(img_vis, ppt, npt, 1, 1, CV_RGB(0, 255, 0), 2, 8, 0);
}
// 调试用，自行替换需要修改的路径
cv::imwrite("/sdcard/1/vis.png", img_vis);
}
std::vector<OCRPredictResult>
OCR_PPredictor::infer_ocr(const std::vector<int64_t> &dims, const float *input_data, int input_len,
int net_flag, cv::Mat &origin) {
PredictorInput input = _det_predictor->get_first_input();
input.set_dims(dims);
input.set_data(input_data, input_len);
std::vector<PredictorOutput> results = _det_predictor->infer();
PredictorOutput &res = results.at(0);
std::vector<std::vector<std::vector<int>>> filtered_box
= calc_filtered_boxes(res.get_float_data(), res.get_size(), (int) dims[2], (int) dims[3],
origin);
LOGI("Filter_box size %ld", filtered_box.size());
return infer_rec(filtered_box, origin);
}
std::vector<OCRPredictResult>
OCR_PPredictor::infer_rec(const std::vector<std::vector<std::vector<int>>> &boxes,
const cv::Mat &origin_img) {
std::vector<float> mean = {0.5f, 0.5f, 0.5f};
std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
std::vector<int64_t> dims = {1, 3, 0, 0};
std::vector<OCRPredictResult> ocr_results;
PredictorInput input = _rec_predictor->get_first_input();
for (auto bp = boxes.crbegin(); bp != boxes.crend(); ++bp) {
const std::vector<std::vector<int>> &box = *bp;
cv::Mat crop_img = get_rotate_crop_image(origin_img, box);
float wh_ratio = float(crop_img.cols) / float(crop_img.rows);
cv::Mat input_image = crnn_resize_img(crop_img, wh_ratio);
input_image.convertTo(input_image, CV_32FC3, 1 / 255.0f);
const float *dimg = reinterpret_cast<const float *>(input_image.data);
int input_size = input_image.rows * input_image.cols;
dims[2] = input_image.rows;
dims[3] = input_image.cols;
input.set_dims(dims);
neon_mean_scale(dimg, input.get_mutable_float_data(), input_size, mean, scale);
std::vector<PredictorOutput> results = _rec_predictor->infer();
OCRPredictResult res;
res.word_index = postprocess_rec_word_index(results.at(0));
if (res.word_index.empty()) {
continue;
}
res.score = postprocess_rec_score(results.at(1));
res.points = box;
ocr_results.emplace_back(std::move(res));
}
LOGI("ocr_results finished %lu", ocr_results.size());
return ocr_results;
}
std::vector<std::vector<std::vector<int>>>
OCR_PPredictor::calc_filtered_boxes(const float *pred, int pred_size, int output_height,
int output_width, const cv::Mat &origin) {
const double threshold = 0.3;
const double maxvalue = 1;
cv::Mat pred_map = cv::Mat::zeros(output_height, output_width, CV_32F);
memcpy(pred_map.data, pred, pred_size * sizeof(float));
cv::Mat cbuf_map;
pred_map.convertTo(cbuf_map, CV_8UC1);
cv::Mat bit_map;
cv::threshold(cbuf_map, bit_map, threshold, maxvalue, cv::THRESH_BINARY);
std::vector<std::vector<std::vector<int>>> boxes = boxes_from_bitmap(pred_map, bit_map);
float ratio_h = output_height * 1.0f / origin.rows;
float ratio_w = output_width * 1.0f / origin.cols;
std::vector<std::vector<std::vector<int>>> filter_boxes = filter_tag_det_res(boxes, ratio_h,
ratio_w, origin);
return filter_boxes;
}
std::vector<int> OCR_PPredictor::postprocess_rec_word_index(const PredictorOutput &res) {
const int *rec_idx = res.get_int_data();
const std::vector<std::vector<uint64_t>> rec_idx_lod = res.get_lod();
std::vector<int> pred_idx;
for (int n = int(rec_idx_lod[0][0]); n < int(rec_idx_lod[0][1] * 2); n += 2) {
pred_idx.emplace_back(rec_idx[n]);
}
return pred_idx;
}
float OCR_PPredictor::postprocess_rec_score(const PredictorOutput &res) {
const float *predict_batch = res.get_float_data();
const std::vector<int64_t> predict_shape = res.get_shape();
const std::vector<std::vector<uint64_t>> predict_lod = res.get_lod();
int blank = predict_shape[1];
float score = 0.f;
int count = 0;
for (int n = predict_lod[0][0]; n < predict_lod[0][1] - 1; n++) {
int argmax_idx = argmax(predict_batch + n * predict_shape[1],
predict_batch + (n + 1) * predict_shape[1]);
float max_value = predict_batch[n * predict_shape[1] + argmax_idx];
if (blank - 1 - argmax_idx > 1e-5) {
score += max_value;
count += 1;
}
}
if (count == 0) {
LOGE("calc score count 0");
} else {
score /= count;
}
LOGI("calc score: %f", score);
return score;
}
NET_TYPE OCR_PPredictor::get_net_flag() const {
return NET_OCR;
}
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/ocr_ppredictor.h 0 → 100644
浏览文件 @ 26219d5f
//
// Created by fujiayi on 2020/7/1.
//
#pragma once
#include <string>
#include <opencv2/opencv.hpp>
#include <paddle_api.h>
#include "ppredictor.h"
namespace ppredictor {
/**
* 配置
*/
struct OCR_Config {
int thread_num = 4; // 线程数
paddle::lite_api::PowerMode mode = paddle::lite_api::LITE_POWER_HIGH; // PaddleLite Mode
};
/**
* 一个四边形内图片的推理结果,
*/
struct OCRPredictResult {
std::vector<int> word_index; //
std::vector<std::vector<int>> points;
float score;
};
/**
* OCR 一共有2个模型进行推理，
* 1. 使用第一个模型（det），框选出多个四边形
* 2. 从原图从抠出这些多边形，使用第二个模型（rec），获取文本
*/
class OCR_PPredictor : public PPredictor_Interface {
public:
OCR_PPredictor(const OCR_Config &config);
virtual ~OCR_PPredictor() {
}
/**
* 初始化二个模型的Predictor
* @param det_model_content
* @param rec_model_content
* @return
*/
int init(const std::string &det_model_content, const std::string &rec_model_content);
int init_from_file(const std::string &det_model_path, const std::string &rec_model_path);
/**
* 返回OCR结果
* @param dims
* @param input_data
* @param input_len
* @param net_flag
* @param origin
* @return
*/
virtual std::vector<OCRPredictResult>
infer_ocr(const std::vector<int64_t> &dims, const float *input_data, int input_len,
int net_flag, cv::Mat &origin);
virtual NET_TYPE get_net_flag() const;
private:
/**
* 从第一个模型的结果中计算有文字的四边形
* @param pred
* @param output_height
* @param output_width
* @param origin
* @return
*/
std::vector<std::vector<std::vector<int>>>
calc_filtered_boxes(const float *pred, int pred_size, int output_height, int output_width,
const cv::Mat &origin);
/**
* 第二个模型的推理
*
* @param boxes
* @param origin
* @return
*/
std::vector<OCRPredictResult>
infer_rec(const std::vector<std::vector<std::vector<int>>> &boxes, const cv::Mat &origin);
/**
* 第二个模型提取文字的后处理
* @param res
* @return
*/
std::vector<int> postprocess_rec_word_index(const PredictorOutput &res);
/**
* 计算第二个模型的文字的置信度
* @param res
* @return
*/
float postprocess_rec_score(const PredictorOutput &res);
std::unique_ptr<PPredictor> _det_predictor;
std::unique_ptr<PPredictor> _rec_predictor;
OCR_Config _config;
};
}
deploy/android_demo/app/src/main/cpp/ppredictor.cpp 0 → 100644
浏览文件 @ 26219d5f
#include "ppredictor.h"
#include "common.h"
namespace ppredictor {
PPredictor::PPredictor(int thread_num, int net_flag, paddle::lite_api::PowerMode mode) :
_thread_num(thread_num), _net_flag(net_flag), _mode(mode) {
}
int PPredictor::init_nb(const std::string &model_content) {
paddle::lite_api::MobileConfig config;
config.set_model_from_buffer(model_content);
return _init(config);
}
int PPredictor::init_from_file(const std::string &model_content){
paddle::lite_api::MobileConfig config;
config.set_model_from_file(model_content);
return _init(config);
}
template<typename ConfigT>
int PPredictor::_init(ConfigT &config) {
config.set_threads(_thread_num);
config.set_power_mode(_mode);
_predictor = paddle::lite_api::CreatePaddlePredictor(config);
LOGI("paddle instance created");
return RETURN_OK;
}
PredictorInput PPredictor::get_input(int index) {
PredictorInput input{_predictor->GetInput(index), index, _net_flag};
_is_input_get = true;
return input;
}
std::vector<PredictorInput> PPredictor::get_inputs(int num) {
std::vector<PredictorInput> results;
for (int i = 0; i < num; i++) {
results.emplace_back(get_input(i));
}
return results;
}
PredictorInput PPredictor::get_first_input() {
return get_input(0);
}
std::vector<PredictorOutput> PPredictor::infer() {
LOGI("infer Run start %d", _net_flag);
std::vector<PredictorOutput> results;
if (!_is_input_get) {
return results;
}
_predictor->Run();
LOGI("infer Run end");
for (int i = 0; i < _predictor->GetOutputNames().size(); i++) {
std::unique_ptr<const paddle::lite_api::Tensor> output_tensor = _predictor->GetOutput(i);
LOGI("output tensor[%d] size %ld", i, product(output_tensor->shape()));
PredictorOutput result{std::move(output_tensor), i, _net_flag};
results.emplace_back(std::move(result));
}
return results;
}
NET_TYPE PPredictor::get_net_flag() const {
return (NET_TYPE) _net_flag;
}
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/ppredictor.h 0 → 100644
浏览文件 @ 26219d5f
#pragma once
#include "paddle_api.h"
#include "predictor_input.h"
#include "predictor_output.h"
namespace ppredictor {
/**
* PaddleLite Preditor 通用接口
*/
class PPredictor_Interface {
public:
virtual ~PPredictor_Interface() {
}
virtual NET_TYPE get_net_flag() const = 0;
};
/**
* 通用推理
*/
class PPredictor : public PPredictor_Interface {
public:
PPredictor(int thread_num, int net_flag = 0,
paddle::lite_api::PowerMode mode = paddle::lite_api::LITE_POWER_HIGH);
virtual ~PPredictor() {
}
/**
* 初始化paddlitelite的opt模型，nb格式，与init_paddle二选一
* @param model_content
* @return 0 目前是固定值0， 之后其他值表示失败
*/
virtual int init_nb(const std::string &model_content);
virtual int init_from_file(const std::string &model_content);
std::vector<PredictorOutput> infer();
std::shared_ptr<paddle::lite_api::PaddlePredictor> get_predictor() {
return _predictor;
}
virtual std::vector<PredictorInput> get_inputs(int num);
virtual PredictorInput get_input(int index);
virtual PredictorInput get_first_input();
virtual NET_TYPE get_net_flag() const;
protected:
template<typename ConfigT>
int _init(ConfigT &config);
private:
int _thread_num;
paddle::lite_api::PowerMode _mode;
std::shared_ptr<paddle::lite_api::PaddlePredictor> _predictor;
bool _is_input_get = false;
int _net_flag;
};
}
deploy/android_demo/app/src/main/cpp/predictor_input.cpp 0 → 100644
浏览文件 @ 26219d5f
#include "predictor_input.h"
namespace ppredictor {
void PredictorInput::set_dims(std::vector<int64_t> dims) {
// yolov3
if (_net_flag == 101 && _index == 1) {
_tensor->Resize({1, 2});
_tensor->mutable_data<int>()[0] = (int) dims.at(2);
_tensor->mutable_data<int>()[1] = (int) dims.at(3);
} else {
_tensor->Resize(dims);
}
_is_dims_set = true;
}
float *PredictorInput::get_mutable_float_data() {
if (!_is_dims_set) {
LOGE("PredictorInput::set_dims is not called");
}
return _tensor->mutable_data<float>();
}
void PredictorInput::set_data(const float *input_data, int input_float_len) {
float *input_raw_data = get_mutable_float_data();
memcpy(input_raw_data, input_data, input_float_len * sizeof(float));
}
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/predictor_input.h 0 → 100644
浏览文件 @ 26219d5f
#pragma once
#include <paddle_api.h>
#include <vector>
#include "common.h"
namespace ppredictor {
class PredictorInput {
public:
PredictorInput(std::unique_ptr<paddle::lite_api::Tensor> &&tensor, int index, int net_flag) :
_tensor(std::move(tensor)), _index(index),_net_flag(net_flag) {
}
void set_dims(std::vector<int64_t> dims);
float *get_mutable_float_data();
void set_data(const float *input_data, int input_float_len);
private:
std::unique_ptr<paddle::lite_api::Tensor> _tensor;
bool _is_dims_set = false;
int _index;
int _net_flag;
};
}
deploy/android_demo/app/src/main/cpp/predictor_output.cpp 0 → 100644
浏览文件 @ 26219d5f
#include "predictor_output.h"
namespace ppredictor {
const float* PredictorOutput::get_float_data() const{
return _tensor->data<float>();
}
const int* PredictorOutput::get_int_data() const{
return _tensor->data<int>();
}
const std::vector<std::vector<uint64_t>> PredictorOutput::get_lod() const{
return _tensor->lod();
}
int64_t PredictorOutput::get_size() const{
if (_net_flag == NET_OCR) {
return _tensor->shape().at(2) * _tensor->shape().at(3);
} else {
return product(_tensor->shape());
}
}
const std::vector<int64_t> PredictorOutput::get_shape() const{
return _tensor->shape();
}
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/predictor_output.h 0 → 100644
浏览文件 @ 26219d5f
#pragma once
#include <paddle_api.h>
#include <vector>
#include "common.h"
namespace ppredictor {
class PredictorOutput {
public:
PredictorOutput(){
}
PredictorOutput(std::unique_ptr<const paddle::lite_api::Tensor> &&tensor, int index, int net_flag) :
_tensor(std::move(tensor)), _index(index), _net_flag(net_flag) {
}
const float* get_float_data() const;
const int* get_int_data() const;
int64_t get_size() const;
const std::vector<std::vector<uint64_t>> get_lod() const;
const std::vector<int64_t> get_shape() const;
std::vector<float> data; // 通常是float返回，与下面的data_int二选一
std::vector<int> data_int; // 少数层是int返回，与 data二选一
std::vector<int64_t> shape; // PaddleLite输出层的shape
std::vector<std::vector<uint64_t>> lod; // PaddleLite输出层的lod
private:
std::unique_ptr<const paddle::lite_api::Tensor> _tensor;
int _index;
int _net_flag;
};
}
deploy/android_demo/app/src/main/cpp/preprocess.cpp 0 → 100644
浏览文件 @ 26219d5f
#include "preprocess.h"
#include <android/bitmap.h>
cv::Mat bitmap_to_cv_mat(JNIEnv *env, jobject bitmap) {
AndroidBitmapInfo info;
int result = AndroidBitmap_getInfo(env, bitmap, &info);
if (result != ANDROID_BITMAP_RESULT_SUCCESS) {
LOGE("AndroidBitmap_getInfo failed, result: %d", result);
return cv::Mat{};
}
if (info.format != ANDROID_BITMAP_FORMAT_RGBA_8888) {
LOGE("Bitmap format is not RGBA_8888 !");
return cv::Mat{};
}
unsigned char *srcData = NULL;
AndroidBitmap_lockPixels(env, bitmap, (void **) &srcData);
cv::Mat mat = cv::Mat::zeros(info.height, info.width, CV_8UC4);
memcpy(mat.data, srcData, info.height * info.width * 4);
AndroidBitmap_unlockPixels(env, bitmap);
cv::cvtColor(mat, mat, cv::COLOR_RGBA2BGR);
/**
if (!cv::imwrite("/sdcard/1/copy.jpg", mat)){
LOGE("Write image failed " );
}
*/
return mat;
}
cv::Mat resize_img(const cv::Mat& img, int height, int width){
if (img.rows == height && img.cols == width){
return img;
}
cv::Mat new_img;
cv::resize(img, new_img, cv::Size(height, width));
return new_img;
}
// fill tensor with mean and scale and trans layout: nhwc -> nchw, neon speed up
void neon_mean_scale(const float* din,
float* dout,
int size,
const std::vector<float>& mean,
const std::vector<float>& scale) {
if (mean.size() != 3 || scale.size() != 3) {
LOGE("[ERROR] mean or scale size must equal to 3");
return;
}
float32x4_t vmean0 = vdupq_n_f32(mean[0]);
float32x4_t vmean1 = vdupq_n_f32(mean[1]);
float32x4_t vmean2 = vdupq_n_f32(mean[2]);
float32x4_t vscale0 = vdupq_n_f32(scale[0]);
float32x4_t vscale1 = vdupq_n_f32(scale[1]);
float32x4_t vscale2 = vdupq_n_f32(scale[2]);
float* dout_c0 = dout;
float* dout_c1 = dout + size;
float* dout_c2 = dout + size * 2;
int i = 0;
for (; i < size - 3; i += 4) {
float32x4x3_t vin3 = vld3q_f32(din);
float32x4_t vsub0 = vsubq_f32(vin3.val[0], vmean0);
float32x4_t vsub1 = vsubq_f32(vin3.val[1], vmean1);
float32x4_t vsub2 = vsubq_f32(vin3.val[2], vmean2);
float32x4_t vs0 = vmulq_f32(vsub0, vscale0);
float32x4_t vs1 = vmulq_f32(vsub1, vscale1);
float32x4_t vs2 = vmulq_f32(vsub2, vscale2);
vst1q_f32(dout_c0, vs0);
vst1q_f32(dout_c1, vs1);
vst1q_f32(dout_c2, vs2);
din += 12;
dout_c0 += 4;
dout_c1 += 4;
dout_c2 += 4;
}
for (; i < size; i++) {
*(dout_c0++) = (*(din++) - mean[0]) * scale[0];
*(dout_c1++) = (*(din++) - mean[1]) * scale[1];
*(dout_c2++) = (*(din++) - mean[2]) * scale[2];
}
}
\ No newline at end of file
deploy/android_demo/app/src/main/cpp/preprocess.h 0 → 100644
浏览文件 @ 26219d5f
#pragma once
#include <jni.h>
#include <opencv2/opencv.hpp>
#include "common.h"
cv::Mat bitmap_to_cv_mat(JNIEnv *env, jobject bitmap);
cv::Mat resize_img(const cv::Mat& img, int height, int width);
void neon_mean_scale(const float* din,
float* dout,
int size,
const std::vector<float>& mean,
const std::vector<float>& scale);
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/AppCompatPreferenceActivity.java 0 → 100644
浏览文件 @ 26219d5f
/*
* Copyright (C) 2014 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.baidu.paddle.lite.demo.ocr;
import android.content.res.Configuration;
import android.os.Bundle;
import android.preference.PreferenceActivity;
import android.support.annotation.LayoutRes;
import android.support.annotation.Nullable;
import android.support.v7.app.ActionBar;
import android.support.v7.app.AppCompatDelegate;
import android.support.v7.widget.Toolbar;
import android.view.MenuInflater;
import android.view.View;
import android.view.ViewGroup;
/**
* A {@link PreferenceActivity} which implements and proxies the necessary calls
* to be used with AppCompat.
* <p>
* This technique can be used with an {@link android.app.Activity} class, not just
* {@link PreferenceActivity}.
*/
public abstract class AppCompatPreferenceActivity extends PreferenceActivity {
private AppCompatDelegate mDelegate;
@Override
protected void onCreate(Bundle savedInstanceState) {
getDelegate().installViewFactory();
getDelegate().onCreate(savedInstanceState);
super.onCreate(savedInstanceState);
}
@Override
protected void onPostCreate(Bundle savedInstanceState) {
super.onPostCreate(savedInstanceState);
getDelegate().onPostCreate(savedInstanceState);
}
public ActionBar getSupportActionBar() {
return getDelegate().getSupportActionBar();
}
public void setSupportActionBar(@Nullable Toolbar toolbar) {
getDelegate().setSupportActionBar(toolbar);
}
@Override
public MenuInflater getMenuInflater() {
return getDelegate().getMenuInflater();
}
@Override
public void setContentView(@LayoutRes int layoutResID) {
getDelegate().setContentView(layoutResID);
}
@Override
public void setContentView(View view) {
getDelegate().setContentView(view);
}
@Override
public void setContentView(View view, ViewGroup.LayoutParams params) {
getDelegate().setContentView(view, params);
}
@Override
public void addContentView(View view, ViewGroup.LayoutParams params) {
getDelegate().addContentView(view, params);
}
@Override
protected void onPostResume() {
super.onPostResume();
getDelegate().onPostResume();
}
@Override
protected void onTitleChanged(CharSequence title, int color) {
super.onTitleChanged(title, color);
getDelegate().setTitle(title);
}
@Override
public void onConfigurationChanged(Configuration newConfig) {
super.onConfigurationChanged(newConfig);
getDelegate().onConfigurationChanged(newConfig);
}
@Override
protected void onStop() {
super.onStop();
getDelegate().onStop();
}
@Override
protected void onDestroy() {
super.onDestroy();
getDelegate().onDestroy();
}
public void invalidateOptionsMenu() {
getDelegate().invalidateOptionsMenu();
}
private AppCompatDelegate getDelegate() {
if (mDelegate == null) {
mDelegate = AppCompatDelegate.create(this, null);
}
return mDelegate;
}
}
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/MainActivity.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import android.Manifest;
import android.app.ProgressDialog;
import android.content.ContentResolver;
import android.content.Intent;
import android.content.SharedPreferences;
import android.content.pm.PackageManager;
import android.database.Cursor;
import android.graphics.Bitmap;
import android.graphics.BitmapFactory;
import android.net.Uri;
import android.os.Bundle;
import android.os.Handler;
import android.os.HandlerThread;
import android.os.Message;
import android.preference.PreferenceManager;
import android.provider.MediaStore;
import android.support.annotation.NonNull;
import android.support.v4.app.ActivityCompat;
import android.support.v4.content.ContextCompat;
import android.support.v7.app.AppCompatActivity;
import android.text.method.ScrollingMovementMethod;
import android.util.Log;
import android.view.Menu;
import android.view.MenuInflater;
import android.view.MenuItem;
import android.widget.ImageView;
import android.widget.TextView;
import android.widget.Toast;
import java.io.File;
import java.io.IOException;
import java.io.InputStream;
public class MainActivity extends AppCompatActivity {
private static final String TAG = MainActivity.class.getSimpleName();
public static final int OPEN_GALLERY_REQUEST_CODE = 0;
public static final int TAKE_PHOTO_REQUEST_CODE = 1;
public static final int REQUEST_LOAD_MODEL = 0;
public static final int REQUEST_RUN_MODEL = 1;
public static final int RESPONSE_LOAD_MODEL_SUCCESSED = 0;
public static final int RESPONSE_LOAD_MODEL_FAILED = 1;
public static final int RESPONSE_RUN_MODEL_SUCCESSED = 2;
public static final int RESPONSE_RUN_MODEL_FAILED = 3;
protected ProgressDialog pbLoadModel = null;
protected ProgressDialog pbRunModel = null;
protected Handler receiver = null; // Receive messages from worker thread
protected Handler sender = null; // Send command to worker thread
protected HandlerThread worker = null; // Worker thread to load&run model
// UI components of object detection
protected TextView tvInputSetting;
protected ImageView ivInputImage;
protected TextView tvOutputResult;
protected TextView tvInferenceTime;
// Model settings of object detection
protected String modelPath = "";
protected String labelPath = "";
protected String imagePath = "";
protected int cpuThreadNum = 1;
protected String cpuPowerMode = "";
protected String inputColorFormat = "";
protected long[] inputShape = new long[]{};
protected float[] inputMean = new float[]{};
protected float[] inputStd = new float[]{};
protected float scoreThreshold = 0.1f;
protected Predictor predictor = new Predictor();
@Override
protected void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
setContentView(R.layout.activity_main);
// Clear all setting items to avoid app crashing due to the incorrect settings
SharedPreferences sharedPreferences = PreferenceManager.getDefaultSharedPreferences(this);
SharedPreferences.Editor editor = sharedPreferences.edit();
editor.clear();
editor.commit();
// Prepare the worker thread for mode loading and inference
receiver = new Handler() {
@Override
public void handleMessage(Message msg) {
switch (msg.what) {
case RESPONSE_LOAD_MODEL_SUCCESSED:
pbLoadModel.dismiss();
onLoadModelSuccessed();
break;
case RESPONSE_LOAD_MODEL_FAILED:
pbLoadModel.dismiss();
Toast.makeText(MainActivity.this, "Load model failed!", Toast.LENGTH_SHORT).show();
onLoadModelFailed();
break;
case RESPONSE_RUN_MODEL_SUCCESSED:
pbRunModel.dismiss();
onRunModelSuccessed();
break;
case RESPONSE_RUN_MODEL_FAILED:
pbRunModel.dismiss();
Toast.makeText(MainActivity.this, "Run model failed!", Toast.LENGTH_SHORT).show();
onRunModelFailed();
break;
default:
break;
}
}
};
worker = new HandlerThread("Predictor Worker");
worker.start();
sender = new Handler(worker.getLooper()) {
public void handleMessage(Message msg) {
switch (msg.what) {
case REQUEST_LOAD_MODEL:
// Load model and reload test image
if (onLoadModel()) {
receiver.sendEmptyMessage(RESPONSE_LOAD_MODEL_SUCCESSED);
} else {
receiver.sendEmptyMessage(RESPONSE_LOAD_MODEL_FAILED);
}
break;
case REQUEST_RUN_MODEL:
// Run model if model is loaded
if (onRunModel()) {
receiver.sendEmptyMessage(RESPONSE_RUN_MODEL_SUCCESSED);
} else {
receiver.sendEmptyMessage(RESPONSE_RUN_MODEL_FAILED);
}
break;
default:
break;
}
}
};
// Setup the UI components
tvInputSetting = findViewById(R.id.tv_input_setting);
ivInputImage = findViewById(R.id.iv_input_image);
tvInferenceTime = findViewById(R.id.tv_inference_time);
tvOutputResult = findViewById(R.id.tv_output_result);
tvInputSetting.setMovementMethod(ScrollingMovementMethod.getInstance());
tvOutputResult.setMovementMethod(ScrollingMovementMethod.getInstance());
}
@Override
protected void onResume() {
super.onResume();
SharedPreferences sharedPreferences = PreferenceManager.getDefaultSharedPreferences(this);
boolean settingsChanged = false;
String model_path = sharedPreferences.getString(getString(R.string.MODEL_PATH_KEY),
getString(R.string.MODEL_PATH_DEFAULT));
String label_path = sharedPreferences.getString(getString(R.string.LABEL_PATH_KEY),
getString(R.string.LABEL_PATH_DEFAULT));
String image_path = sharedPreferences.getString(getString(R.string.IMAGE_PATH_KEY),
getString(R.string.IMAGE_PATH_DEFAULT));
settingsChanged |= !model_path.equalsIgnoreCase(modelPath);
settingsChanged |= !label_path.equalsIgnoreCase(labelPath);
settingsChanged |= !image_path.equalsIgnoreCase(imagePath);
int cpu_thread_num = Integer.parseInt(sharedPreferences.getString(getString(R.string.CPU_THREAD_NUM_KEY),
getString(R.string.CPU_THREAD_NUM_DEFAULT)));
settingsChanged |= cpu_thread_num != cpuThreadNum;
String cpu_power_mode =
sharedPreferences.getString(getString(R.string.CPU_POWER_MODE_KEY),
getString(R.string.CPU_POWER_MODE_DEFAULT));
settingsChanged |= !cpu_power_mode.equalsIgnoreCase(cpuPowerMode);
String input_color_format =
sharedPreferences.getString(getString(R.string.INPUT_COLOR_FORMAT_KEY),
getString(R.string.INPUT_COLOR_FORMAT_DEFAULT));
settingsChanged |= !input_color_format.equalsIgnoreCase(inputColorFormat);
long[] input_shape =
Utils.parseLongsFromString(sharedPreferences.getString(getString(R.string.INPUT_SHAPE_KEY),
getString(R.string.INPUT_SHAPE_DEFAULT)), ",");
float[] input_mean =
Utils.parseFloatsFromString(sharedPreferences.getString(getString(R.string.INPUT_MEAN_KEY),
getString(R.string.INPUT_MEAN_DEFAULT)), ",");
float[] input_std =
Utils.parseFloatsFromString(sharedPreferences.getString(getString(R.string.INPUT_STD_KEY)
, getString(R.string.INPUT_STD_DEFAULT)), ",");
settingsChanged |= input_shape.length != inputShape.length;
settingsChanged |= input_mean.length != inputMean.length;
settingsChanged |= input_std.length != inputStd.length;
if (!settingsChanged) {
for (int i = 0; i < input_shape.length; i++) {
settingsChanged |= input_shape[i] != inputShape[i];
}
for (int i = 0; i < input_mean.length; i++) {
settingsChanged |= input_mean[i] != inputMean[i];
}
for (int i = 0; i < input_std.length; i++) {
settingsChanged |= input_std[i] != inputStd[i];
}
}
float score_threshold =
Float.parseFloat(sharedPreferences.getString(getString(R.string.SCORE_THRESHOLD_KEY),
getString(R.string.SCORE_THRESHOLD_DEFAULT)));
settingsChanged |= scoreThreshold != score_threshold;
if (settingsChanged) {
modelPath = model_path;
labelPath = label_path;
imagePath = image_path;
cpuThreadNum = cpu_thread_num;
cpuPowerMode = cpu_power_mode;
inputColorFormat = input_color_format;
inputShape = input_shape;
inputMean = input_mean;
inputStd = input_std;
scoreThreshold = score_threshold;
// Update UI
tvInputSetting.setText("Model: " + modelPath.substring(modelPath.lastIndexOf("/") + 1) + "\n" + "CPU" +
" Thread Num: " + Integer.toString(cpuThreadNum) + "\n" + "CPU Power Mode: " + cpuPowerMode);
tvInputSetting.scrollTo(0, 0);
// Reload model if configure has been changed
loadModel();
}
}
public void loadModel() {
pbLoadModel = ProgressDialog.show(this, "", "Loading model...", false, false);
sender.sendEmptyMessage(REQUEST_LOAD_MODEL);
}
public void runModel() {
pbRunModel = ProgressDialog.show(this, "", "Running model...", false, false);
sender.sendEmptyMessage(REQUEST_RUN_MODEL);
}
public boolean onLoadModel() {
return predictor.init(MainActivity.this, modelPath, labelPath, cpuThreadNum,
cpuPowerMode,
inputColorFormat,
inputShape, inputMean,
inputStd, scoreThreshold);
}
public boolean onRunModel() {
return predictor.isLoaded() && predictor.runModel();
}
public void onLoadModelSuccessed() {
// Load test image from path and run model
try {
if (imagePath.isEmpty()) {
return;
}
Bitmap image = null;
// Read test image file from custom path if the first character of mode path is '/', otherwise read test
// image file from assets
if (!imagePath.substring(0, 1).equals("/")) {
InputStream imageStream = getAssets().open(imagePath);
image = BitmapFactory.decodeStream(imageStream);
} else {
if (!new File(imagePath).exists()) {
return;
}
image = BitmapFactory.decodeFile(imagePath);
}
if (image != null && predictor.isLoaded()) {
predictor.setInputImage(image);
runModel();
}
} catch (IOException e) {
Toast.makeText(MainActivity.this, "Load image failed!", Toast.LENGTH_SHORT).show();
e.printStackTrace();
}
}
public void onLoadModelFailed() {
}
public void onRunModelSuccessed() {
// Obtain results and update UI
tvInferenceTime.setText("Inference time: " + predictor.inferenceTime() + " ms");
Bitmap outputImage = predictor.outputImage();
if (outputImage != null) {
ivInputImage.setImageBitmap(outputImage);
}
tvOutputResult.setText(predictor.outputResult());
tvOutputResult.scrollTo(0, 0);
}
public void onRunModelFailed() {
}
public void onImageChanged(Bitmap image) {
// Rerun model if users pick test image from gallery or camera
if (image != null && predictor.isLoaded()) {
predictor.setInputImage(image);
runModel();
}
}
public void onSettingsClicked() {
startActivity(new Intent(MainActivity.this, SettingsActivity.class));
}
@Override
public boolean onCreateOptionsMenu(Menu menu) {
MenuInflater inflater = getMenuInflater();
inflater.inflate(R.menu.menu_action_options, menu);
return true;
}
public boolean onPrepareOptionsMenu(Menu menu) {
boolean isLoaded = predictor.isLoaded();
menu.findItem(R.id.open_gallery).setEnabled(isLoaded);
menu.findItem(R.id.take_photo).setEnabled(isLoaded);
return super.onPrepareOptionsMenu(menu);
}
@Override
public boolean onOptionsItemSelected(MenuItem item) {
switch (item.getItemId()) {
case android.R.id.home:
finish();
break;
case R.id.open_gallery:
if (requestAllPermissions()) {
openGallery();
}
break;
case R.id.take_photo:
if (requestAllPermissions()) {
takePhoto();
}
break;
case R.id.settings:
if (requestAllPermissions()) {
// Make sure we have SDCard r&w permissions to load model from SDCard
onSettingsClicked();
}
break;
}
return super.onOptionsItemSelected(item);
}
@Override
public void onRequestPermissionsResult(int requestCode, @NonNull String[] permissions,
@NonNull int[] grantResults) {
super.onRequestPermissionsResult(requestCode, permissions, grantResults);
if (grantResults[0] != PackageManager.PERMISSION_GRANTED || grantResults[1] != PackageManager.PERMISSION_GRANTED) {
Toast.makeText(this, "Permission Denied", Toast.LENGTH_SHORT).show();
}
}
private boolean requestAllPermissions() {
if (ContextCompat.checkSelfPermission(this, Manifest.permission.WRITE_EXTERNAL_STORAGE)
!= PackageManager.PERMISSION_GRANTED || ContextCompat.checkSelfPermission(this,
Manifest.permission.CAMERA)
!= PackageManager.PERMISSION_GRANTED) {
ActivityCompat.requestPermissions(this, new String[]{Manifest.permission.WRITE_EXTERNAL_STORAGE,
Manifest.permission.CAMERA},
0);
return false;
}
return true;
}
private void openGallery() {
Intent intent = new Intent(Intent.ACTION_PICK, null);
intent.setDataAndType(MediaStore.Images.Media.EXTERNAL_CONTENT_URI, "image/*");
startActivityForResult(intent, OPEN_GALLERY_REQUEST_CODE);
}
private void takePhoto() {
Intent takePhotoIntent = new Intent(MediaStore.ACTION_IMAGE_CAPTURE);
if (takePhotoIntent.resolveActivity(getPackageManager()) != null) {
startActivityForResult(takePhotoIntent, TAKE_PHOTO_REQUEST_CODE);
}
}
@Override
protected void onActivityResult(int requestCode, int resultCode, Intent data) {
super.onActivityResult(requestCode, resultCode, data);
if (resultCode == RESULT_OK && data != null) {
switch (requestCode) {
case OPEN_GALLERY_REQUEST_CODE:
try {
ContentResolver resolver = getContentResolver();
Uri uri = data.getData();
Bitmap image = MediaStore.Images.Media.getBitmap(resolver, uri);
String[] proj = {MediaStore.Images.Media.DATA};
Cursor cursor = managedQuery(uri, proj, null, null, null);
cursor.moveToFirst();
onImageChanged(image);
} catch (IOException e) {
Log.e(TAG, e.toString());
}
break;
case TAKE_PHOTO_REQUEST_CODE:
Bundle extras = data.getExtras();
Bitmap image = (Bitmap) extras.get("data");
onImageChanged(image);
break;
default:
break;
}
}
}
@Override
protected void onDestroy() {
if (predictor != null) {
predictor.releaseModel();
}
worker.quit();
super.onDestroy();
}
}
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OCRPredictorNative.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import android.graphics.Bitmap;
import android.util.Log;
import java.util.ArrayList;
import java.util.concurrent.atomic.AtomicBoolean;
public class OCRPredictorNative {
private static final AtomicBoolean isSOLoaded = new AtomicBoolean();
public static void loadLibrary() throws RuntimeException {
if (!isSOLoaded.get() && isSOLoaded.compareAndSet(false, true)) {
try {
System.loadLibrary("Native");
} catch (Throwable e) {
RuntimeException exception = new RuntimeException(
"Load libNative.so failed, please check it exists in apk file.", e);
throw exception;
}
}
}
private Config config;
private long nativePointer = 0;
public OCRPredictorNative(Config config) {
this.config = config;
loadLibrary();
nativePointer = init(config.detModelFilename, config.recModelFilename,
config.cpuThreadNum, config.cpuPower);
Log.i("OCRPredictorNative", "load success " + nativePointer);
}
public void release(){
if (nativePointer != 0){
nativePointer = 0;
destory(nativePointer);
}
}
public ArrayList<OcrResultModel> runImage(float[] inputData, int width, int height, int channels, Bitmap originalImage) {
Log.i("OCRPredictorNative", "begin to run image " + inputData.length + " " + width + " " + height);
float[] dims = new float[]{1, channels, height, width};
float[] rawResults = forward(nativePointer, inputData, dims, originalImage);
ArrayList<OcrResultModel> results = postprocess(rawResults);
return results;
}
public static class Config {
public int cpuThreadNum;
public String cpuPower;
public String detModelFilename;
public String recModelFilename;
}
protected native long init(String detModelPath, String recModelPath, int threadNum, String cpuMode);
protected native float[] forward(long pointer, float[] buf, float[] ddims, Bitmap originalImage);
protected native void destory(long pointer);
private ArrayList<OcrResultModel> postprocess(float[] raw) {
ArrayList<OcrResultModel> results = new ArrayList<OcrResultModel>();
int begin = 0;
while (begin < raw.length) {
int point_num = Math.round(raw[begin]);
int word_num = Math.round(raw[begin + 1]);
OcrResultModel model = parse(raw, begin + 2, point_num, word_num);
begin += 2 + 1 + point_num * 2 + word_num;
results.add(model);
}
return results;
}
private OcrResultModel parse(float[] raw, int begin, int pointNum, int wordNum) {
int current = begin;
OcrResultModel model = new OcrResultModel();
model.setConfidence(raw[current]);
current++;
for (int i = 0; i < pointNum; i++) {
model.addPoints(Math.round(raw[current + i * 2]), Math.round(raw[current + i * 2 + 1]));
}
current += (pointNum * 2);
for (int i = 0; i < wordNum; i++) {
int index = Math.round(raw[current + i]);
model.addWordIndex(index);
}
Log.i("OCRPredictorNative", "word finished " + wordNum);
return model;
}
}
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OcrResultModel.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import android.graphics.Point;
import java.util.ArrayList;
import java.util.List;
public class OcrResultModel {
private List<Point> points;
private List<Integer> wordIndex;
private String label;
private float confidence;
public OcrResultModel() {
super();
points = new ArrayList<>();
wordIndex = new ArrayList<>();
}
public void addPoints(int x, int y) {
Point point = new Point(x, y);
points.add(point);
}
public void addWordIndex(int index) {
wordIndex.add(index);
}
public List<Point> getPoints() {
return points;
}
public List<Integer> getWordIndex() {
return wordIndex;
}
public String getLabel() {
return label;
}
public void setLabel(String label) {
this.label = label;
}
public float getConfidence() {
return confidence;
}
public void setConfidence(float confidence) {
this.confidence = confidence;
}
}
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import android.content.Context;
import android.graphics.Bitmap;
import android.graphics.Canvas;
import android.graphics.Color;
import android.graphics.Paint;
import android.graphics.Path;
import android.graphics.Point;
import android.util.Log;
import java.io.File;
import java.io.InputStream;
import java.util.ArrayList;
import java.util.Date;
import java.util.List;
import java.util.Vector;
import static android.graphics.Color.*;
public class Predictor {
private static final String TAG = Predictor.class.getSimpleName();
public boolean isLoaded = false;
public int warmupIterNum = 1;
public int inferIterNum = 1;
public int cpuThreadNum = 4;
public String cpuPowerMode = "LITE_POWER_HIGH";
public String modelPath = "";
public String modelName = "";
protected OCRPredictorNative paddlePredictor = null;
protected float inferenceTime = 0;
// Only for object detection
protected Vector<String> wordLabels = new Vector<String>();
protected String inputColorFormat = "BGR";
protected long[] inputShape = new long[]{1, 3, 960};
protected float[] inputMean = new float[]{0.485f, 0.456f, 0.406f};
protected float[] inputStd = new float[]{1.0f / 0.229f, 1.0f / 0.224f, 1.0f / 0.225f};
protected float scoreThreshold = 0.1f;
protected Bitmap inputImage = null;
protected Bitmap outputImage = null;
protected String outputResult = "";
protected float preprocessTime = 0;
protected float postprocessTime = 0;
public Predictor() {
}
public boolean init(Context appCtx, String modelPath, String labelPath, int cpuThreadNum, String cpuPowerMode,
String inputColorFormat,
long[] inputShape, float[] inputMean,
float[] inputStd, float scoreThreshold) {
if (inputShape.length != 3) {
Log.e(TAG, "Size of input shape should be: 3");
return false;
}
if (inputMean.length != inputShape[1]) {
Log.e(TAG, "Size of input mean should be: " + Long.toString(inputShape[1]));
return false;
}
if (inputStd.length != inputShape[1]) {
Log.e(TAG, "Size of input std should be: " + Long.toString(inputShape[1]));
return false;
}
if (inputShape[0] != 1) {
Log.e(TAG, "Only one batch is supported in the image classification demo, you can use any batch size in " +
"your Apps!");
return false;
}
if (inputShape[1] != 1 && inputShape[1] != 3) {
Log.e(TAG, "Only one/three channels are supported in the image classification demo, you can use any " +
"channel size in your Apps!");
return false;
}
if (!inputColorFormat.equalsIgnoreCase("BGR")) {
Log.e(TAG, "Only BGR color format is supported.");
return false;
}
isLoaded = loadModel(appCtx, modelPath, cpuThreadNum, cpuPowerMode);
if (!isLoaded) {
return false;
}
isLoaded = loadLabel(appCtx, labelPath);
if (!isLoaded) {
return false;
}
this.inputColorFormat = inputColorFormat;
this.inputShape = inputShape;
this.inputMean = inputMean;
this.inputStd = inputStd;
this.scoreThreshold = scoreThreshold;
return true;
}
protected boolean loadModel(Context appCtx, String modelPath, int cpuThreadNum, String cpuPowerMode) {
// Release model if exists
releaseModel();
// Load model
if (modelPath.isEmpty()) {
return false;
}
String realPath = modelPath;
if (!modelPath.substring(0, 1).equals("/")) {
// Read model files from custom path if the first character of mode path is '/'
// otherwise copy model to cache from assets
realPath = appCtx.getCacheDir() + "/" + modelPath;
Utils.copyDirectoryFromAssets(appCtx, modelPath, realPath);
}
if (realPath.isEmpty()) {
return false;
}
OCRPredictorNative.Config config = new OCRPredictorNative.Config();
config.cpuThreadNum = cpuThreadNum;
config.detModelFilename = realPath + File.separator + "ch_det_mv3_db_opt.nb";
config.recModelFilename = realPath + File.separator + "ch_rec_mv3_crnn_opt.nb";
Log.e("Predictor", "model path" + config.detModelFilename + " ; " + config.recModelFilename);
config.cpuPower = cpuPowerMode;
paddlePredictor = new OCRPredictorNative(config);
this.cpuThreadNum = cpuThreadNum;
this.cpuPowerMode = cpuPowerMode;
this.modelPath = realPath;
this.modelName = realPath.substring(realPath.lastIndexOf("/") + 1);
return true;
}
public void releaseModel() {
if (paddlePredictor != null){
paddlePredictor.release();
paddlePredictor = null;
}
isLoaded = false;
cpuThreadNum = 4;
cpuPowerMode = "LITE_POWER_HIGH";
modelPath = "";
modelName = "";
}
protected boolean loadLabel(Context appCtx, String labelPath) {
wordLabels.clear();
// Load word labels from file
try {
InputStream assetsInputStream = appCtx.getAssets().open(labelPath);
int available = assetsInputStream.available();
byte[] lines = new byte[available];
assetsInputStream.read(lines);
assetsInputStream.close();
String words = new String(lines);
String[] contents = words.split("\n");
for (String content : contents) {
wordLabels.add(content);
}
Log.i(TAG, "Word label size: " + wordLabels.size());
} catch (Exception e) {
Log.e(TAG, e.getMessage());
return false;
}
return true;
}
public boolean runModel() {
if (inputImage == null || !isLoaded()) {
return false;
}
// Pre-process image, and feed input tensor with pre-processed data
Bitmap scaleImage = Utils.resizeWithStep(inputImage, Long.valueOf(inputShape[2]).intValue(), 32);
Date start = new Date();
int channels = (int) inputShape[1];
int width = scaleImage.getWidth();
int height = scaleImage.getHeight();
float[] inputData = new float[channels * width * height];
if (channels == 3) {
int[] channelIdx = null;
if (inputColorFormat.equalsIgnoreCase("RGB")) {
channelIdx = new int[]{0, 1, 2};
} else if (inputColorFormat.equalsIgnoreCase("BGR")) {
channelIdx = new int[]{2, 1, 0};
} else {
Log.i(TAG, "Unknown color format " + inputColorFormat + ", only RGB and BGR color format is " +
"supported!");
return false;
}
int[] channelStride = new int[]{width * height, width * height * 2};
int p = scaleImage.getPixel(scaleImage.getWidth() - 1, scaleImage.getHeight() - 1);
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
int color = scaleImage.getPixel(x, y);
float[] rgb = new float[]{(float) red(color) / 255.0f, (float) green(color) / 255.0f,
(float) blue(color) / 255.0f};
inputData[y * width + x] = (rgb[channelIdx[0]] - inputMean[0]) / inputStd[0];
inputData[y * width + x + channelStride[0]] = (rgb[channelIdx[1]] - inputMean[1]) / inputStd[1];
inputData[y * width + x + channelStride[1]] = (rgb[channelIdx[2]] - inputMean[2]) / inputStd[2];
}
}
} else if (channels == 1) {
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
int color = inputImage.getPixel(x, y);
float gray = (float) (red(color) + green(color) + blue(color)) / 3.0f / 255.0f;
inputData[y * width + x] = (gray - inputMean[0]) / inputStd[0];
}
}
} else {
Log.i(TAG, "Unsupported channel size " + Integer.toString(channels) + ", only channel 1 and 3 is " +
"supported!");
return false;
}
float[] pixels = inputData;
Log.i(TAG, "pixels " + pixels[0] + " " + pixels[1] + " " + pixels[2] + " " + pixels[3]
+ " " + pixels[pixels.length / 2] + " " + pixels[pixels.length / 2 + 1] + " " + pixels[pixels.length - 2] + " " + pixels[pixels.length - 1]);
Date end = new Date();
preprocessTime = (float) (end.getTime() - start.getTime());
// Warm up
for (int i = 0; i < warmupIterNum; i++) {
paddlePredictor.runImage(inputData, width, height, channels, inputImage);
}
warmupIterNum = 0; // 之后不要再warm了
// Run inference
start = new Date();
ArrayList<OcrResultModel> results = paddlePredictor.runImage(inputData, width, height, channels, inputImage);
end = new Date();
inferenceTime = (end.getTime() - start.getTime()) / (float) inferIterNum;
results = postprocess(results);
Log.i(TAG, "[stat] Preprocess Time: " + preprocessTime
+ " ; Inference Time: " + inferenceTime + " ;Box Size " + results.size());
drawResults(results);
return true;
}
public boolean isLoaded() {
return paddlePredictor != null && isLoaded;
}
public String modelPath() {
return modelPath;
}
public String modelName() {
return modelName;
}
public int cpuThreadNum() {
return cpuThreadNum;
}
public String cpuPowerMode() {
return cpuPowerMode;
}
public float inferenceTime() {
return inferenceTime;
}
public Bitmap inputImage() {
return inputImage;
}
public Bitmap outputImage() {
return outputImage;
}
public String outputResult() {
return outputResult;
}
public float preprocessTime() {
return preprocessTime;
}
public float postprocessTime() {
return postprocessTime;
}
public void setInputImage(Bitmap image) {
if (image == null) {
return;
}
// Scale image to the size of input tensor
Bitmap rgbaImage = image.copy(Bitmap.Config.ARGB_8888, true);
this.inputImage = rgbaImage;
}
private ArrayList<OcrResultModel> postprocess(ArrayList<OcrResultModel> results) {
for (OcrResultModel r : results) {
StringBuffer word = new StringBuffer();
for (int index : r.getWordIndex()) {
if (index >= 0 && index < wordLabels.size()) {
word.append(wordLabels.get(index));
} else {
Log.e(TAG, "Word index is not in label list:" + index);
word.append("×");
}
}
r.setLabel(word.toString());
}
return results;
}
private void drawResults(ArrayList<OcrResultModel> results) {
StringBuffer outputResultSb = new StringBuffer("");
for (int i=0;i<results.size();i++) {
OcrResultModel result = results.get(i);
StringBuilder sb = new StringBuilder("");
sb.append(result.getLabel());
sb.append(" ").append(result.getConfidence());
sb.append("; Points: ");
for (Point p : result.getPoints()) {
sb.append("(").append(p.x).append(",").append(p.y).append(") ");
}
Log.i(TAG, sb.toString());
outputResultSb.append(i+1).append(": ").append(result.getLabel()).append("\n");
}
outputResult = outputResultSb.toString();
outputImage = inputImage;
Canvas canvas = new Canvas(outputImage);
Paint paintFillAlpha = new Paint();
paintFillAlpha.setStyle(Paint.Style.FILL);
paintFillAlpha.setColor(Color.parseColor("#3B85F5"));
paintFillAlpha.setAlpha(50);
Paint paint = new Paint();
paint.setColor(Color.parseColor("#3B85F5"));
paint.setStrokeWidth(5);
paint.setStyle(Paint.Style.STROKE);
for (OcrResultModel result : results) {
Path path = new Path();
List<Point> points = result.getPoints();
path.moveTo(points.get(0).x, points.get(0).y);
for (int i = points.size() - 1; i >= 0; i--) {
Point p = points.get(i);
path.lineTo(p.x, p.y);
}
canvas.drawPath(path, paint);
canvas.drawPath(path, paintFillAlpha);
}
}
}
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/SettingsActivity.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import android.content.SharedPreferences;
import android.os.Bundle;
import android.preference.CheckBoxPreference;
import android.preference.EditTextPreference;
import android.preference.ListPreference;
import android.support.v7.app.ActionBar;
import java.util.ArrayList;
import java.util.List;
public class SettingsActivity extends AppCompatPreferenceActivity implements SharedPreferences.OnSharedPreferenceChangeListener {
ListPreference lpChoosePreInstalledModel = null;
CheckBoxPreference cbEnableCustomSettings = null;
EditTextPreference etModelPath = null;
EditTextPreference etLabelPath = null;
EditTextPreference etImagePath = null;
ListPreference lpCPUThreadNum = null;
ListPreference lpCPUPowerMode = null;
ListPreference lpInputColorFormat = null;
EditTextPreference etInputShape = null;
EditTextPreference etInputMean = null;
EditTextPreference etInputStd = null;
EditTextPreference etScoreThreshold = null;
List<String> preInstalledModelPaths = null;
List<String> preInstalledLabelPaths = null;
List<String> preInstalledImagePaths = null;
List<String> preInstalledInputShapes = null;
List<String> preInstalledCPUThreadNums = null;
List<String> preInstalledCPUPowerModes = null;
List<String> preInstalledInputColorFormats = null;
List<String> preInstalledInputMeans = null;
List<String> preInstalledInputStds = null;
List<String> preInstalledScoreThresholds = null;
@Override
public void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
addPreferencesFromResource(R.xml.settings);
ActionBar supportActionBar = getSupportActionBar();
if (supportActionBar != null) {
supportActionBar.setDisplayHomeAsUpEnabled(true);
}
// Initialized pre-installed models
preInstalledModelPaths = new ArrayList<String>();
preInstalledLabelPaths = new ArrayList<String>();
preInstalledImagePaths = new ArrayList<String>();
preInstalledInputShapes = new ArrayList<String>();
preInstalledCPUThreadNums = new ArrayList<String>();
preInstalledCPUPowerModes = new ArrayList<String>();
preInstalledInputColorFormats = new ArrayList<String>();
preInstalledInputMeans = new ArrayList<String>();
preInstalledInputStds = new ArrayList<String>();
preInstalledScoreThresholds = new ArrayList<String>();
// Add ssd_mobilenet_v1_pascalvoc_for_cpu
preInstalledModelPaths.add(getString(R.string.MODEL_PATH_DEFAULT));
preInstalledLabelPaths.add(getString(R.string.LABEL_PATH_DEFAULT));
preInstalledImagePaths.add(getString(R.string.IMAGE_PATH_DEFAULT));
preInstalledCPUThreadNums.add(getString(R.string.CPU_THREAD_NUM_DEFAULT));
preInstalledCPUPowerModes.add(getString(R.string.CPU_POWER_MODE_DEFAULT));
preInstalledInputColorFormats.add(getString(R.string.INPUT_COLOR_FORMAT_DEFAULT));
preInstalledInputShapes.add(getString(R.string.INPUT_SHAPE_DEFAULT));
preInstalledInputMeans.add(getString(R.string.INPUT_MEAN_DEFAULT));
preInstalledInputStds.add(getString(R.string.INPUT_STD_DEFAULT));
preInstalledScoreThresholds.add(getString(R.string.SCORE_THRESHOLD_DEFAULT));
// Setup UI components
lpChoosePreInstalledModel =
(ListPreference) findPreference(getString(R.string.CHOOSE_PRE_INSTALLED_MODEL_KEY));
String[] preInstalledModelNames = new String[preInstalledModelPaths.size()];
for (int i = 0; i < preInstalledModelPaths.size(); i++) {
preInstalledModelNames[i] =
preInstalledModelPaths.get(i).substring(preInstalledModelPaths.get(i).lastIndexOf("/") + 1);
}
lpChoosePreInstalledModel.setEntries(preInstalledModelNames);
lpChoosePreInstalledModel.setEntryValues(preInstalledModelPaths.toArray(new String[preInstalledModelPaths.size()]));
cbEnableCustomSettings =
(CheckBoxPreference) findPreference(getString(R.string.ENABLE_CUSTOM_SETTINGS_KEY));
etModelPath = (EditTextPreference) findPreference(getString(R.string.MODEL_PATH_KEY));
etModelPath.setTitle("Model Path (SDCard: " + Utils.getSDCardDirectory() + ")");
etLabelPath = (EditTextPreference) findPreference(getString(R.string.LABEL_PATH_KEY));
etImagePath = (EditTextPreference) findPreference(getString(R.string.IMAGE_PATH_KEY));
lpCPUThreadNum =
(ListPreference) findPreference(getString(R.string.CPU_THREAD_NUM_KEY));
lpCPUPowerMode =
(ListPreference) findPreference(getString(R.string.CPU_POWER_MODE_KEY));
lpInputColorFormat =
(ListPreference) findPreference(getString(R.string.INPUT_COLOR_FORMAT_KEY));
etInputShape = (EditTextPreference) findPreference(getString(R.string.INPUT_SHAPE_KEY));
etInputMean = (EditTextPreference) findPreference(getString(R.string.INPUT_MEAN_KEY));
etInputStd = (EditTextPreference) findPreference(getString(R.string.INPUT_STD_KEY));
etScoreThreshold = (EditTextPreference) findPreference(getString(R.string.SCORE_THRESHOLD_KEY));
}
private void reloadPreferenceAndUpdateUI() {
SharedPreferences sharedPreferences = getPreferenceScreen().getSharedPreferences();
boolean enableCustomSettings =
sharedPreferences.getBoolean(getString(R.string.ENABLE_CUSTOM_SETTINGS_KEY), false);
String modelPath = sharedPreferences.getString(getString(R.string.CHOOSE_PRE_INSTALLED_MODEL_KEY),
getString(R.string.MODEL_PATH_DEFAULT));
int modelIdx = lpChoosePreInstalledModel.findIndexOfValue(modelPath);
if (modelIdx >= 0 && modelIdx < preInstalledModelPaths.size()) {
if (!enableCustomSettings) {
SharedPreferences.Editor editor = sharedPreferences.edit();
editor.putString(getString(R.string.MODEL_PATH_KEY), preInstalledModelPaths.get(modelIdx));
editor.putString(getString(R.string.LABEL_PATH_KEY), preInstalledLabelPaths.get(modelIdx));
editor.putString(getString(R.string.IMAGE_PATH_KEY), preInstalledImagePaths.get(modelIdx));
editor.putString(getString(R.string.CPU_THREAD_NUM_KEY), preInstalledCPUThreadNums.get(modelIdx));
editor.putString(getString(R.string.CPU_POWER_MODE_KEY), preInstalledCPUPowerModes.get(modelIdx));
editor.putString(getString(R.string.INPUT_COLOR_FORMAT_KEY),
preInstalledInputColorFormats.get(modelIdx));
editor.putString(getString(R.string.INPUT_SHAPE_KEY), preInstalledInputShapes.get(modelIdx));
editor.putString(getString(R.string.INPUT_MEAN_KEY), preInstalledInputMeans.get(modelIdx));
editor.putString(getString(R.string.INPUT_STD_KEY), preInstalledInputStds.get(modelIdx));
editor.putString(getString(R.string.SCORE_THRESHOLD_KEY),
preInstalledScoreThresholds.get(modelIdx));
editor.commit();
}
lpChoosePreInstalledModel.setSummary(modelPath);
}
cbEnableCustomSettings.setChecked(enableCustomSettings);
etModelPath.setEnabled(enableCustomSettings);
etLabelPath.setEnabled(enableCustomSettings);
etImagePath.setEnabled(enableCustomSettings);
lpCPUThreadNum.setEnabled(enableCustomSettings);
lpCPUPowerMode.setEnabled(enableCustomSettings);
lpInputColorFormat.setEnabled(enableCustomSettings);
etInputShape.setEnabled(enableCustomSettings);
etInputMean.setEnabled(enableCustomSettings);
etInputStd.setEnabled(enableCustomSettings);
etScoreThreshold.setEnabled(enableCustomSettings);
modelPath = sharedPreferences.getString(getString(R.string.MODEL_PATH_KEY),
getString(R.string.MODEL_PATH_DEFAULT));
String labelPath = sharedPreferences.getString(getString(R.string.LABEL_PATH_KEY),
getString(R.string.LABEL_PATH_DEFAULT));
String imagePath = sharedPreferences.getString(getString(R.string.IMAGE_PATH_KEY),
getString(R.string.IMAGE_PATH_DEFAULT));
String cpuThreadNum = sharedPreferences.getString(getString(R.string.CPU_THREAD_NUM_KEY),
getString(R.string.CPU_THREAD_NUM_DEFAULT));
String cpuPowerMode = sharedPreferences.getString(getString(R.string.CPU_POWER_MODE_KEY),
getString(R.string.CPU_POWER_MODE_DEFAULT));
String inputColorFormat = sharedPreferences.getString(getString(R.string.INPUT_COLOR_FORMAT_KEY),
getString(R.string.INPUT_COLOR_FORMAT_DEFAULT));
String inputShape = sharedPreferences.getString(getString(R.string.INPUT_SHAPE_KEY),
getString(R.string.INPUT_SHAPE_DEFAULT));
String inputMean = sharedPreferences.getString(getString(R.string.INPUT_MEAN_KEY),
getString(R.string.INPUT_MEAN_DEFAULT));
String inputStd = sharedPreferences.getString(getString(R.string.INPUT_STD_KEY),
getString(R.string.INPUT_STD_DEFAULT));
String scoreThreshold = sharedPreferences.getString(getString(R.string.SCORE_THRESHOLD_KEY),
getString(R.string.SCORE_THRESHOLD_DEFAULT));
etModelPath.setSummary(modelPath);
etModelPath.setText(modelPath);
etLabelPath.setSummary(labelPath);
etLabelPath.setText(labelPath);
etImagePath.setSummary(imagePath);
etImagePath.setText(imagePath);
lpCPUThreadNum.setValue(cpuThreadNum);
lpCPUThreadNum.setSummary(cpuThreadNum);
lpCPUPowerMode.setValue(cpuPowerMode);
lpCPUPowerMode.setSummary(cpuPowerMode);
lpInputColorFormat.setValue(inputColorFormat);
lpInputColorFormat.setSummary(inputColorFormat);
etInputShape.setSummary(inputShape);
etInputShape.setText(inputShape);
etInputMean.setSummary(inputMean);
etInputMean.setText(inputMean);
etInputStd.setSummary(inputStd);
etInputStd.setText(inputStd);
etScoreThreshold.setText(scoreThreshold);
etScoreThreshold.setSummary(scoreThreshold);
}
@Override
protected void onResume() {
super.onResume();
getPreferenceScreen().getSharedPreferences().registerOnSharedPreferenceChangeListener(this);
reloadPreferenceAndUpdateUI();
}
@Override
protected void onPause() {
super.onPause();
getPreferenceScreen().getSharedPreferences().unregisterOnSharedPreferenceChangeListener(this);
}
@Override
public void onSharedPreferenceChanged(SharedPreferences sharedPreferences, String key) {
if (key.equals(getString(R.string.CHOOSE_PRE_INSTALLED_MODEL_KEY))) {
SharedPreferences.Editor editor = sharedPreferences.edit();
editor.putBoolean(getString(R.string.ENABLE_CUSTOM_SETTINGS_KEY), false);
editor.commit();
}
reloadPreferenceAndUpdateUI();
}
}
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Utils.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import android.content.Context;
import android.graphics.Bitmap;
import android.os.Environment;
import java.io.*;
public class Utils {
private static final String TAG = Utils.class.getSimpleName();
public static void copyFileFromAssets(Context appCtx, String srcPath, String dstPath) {
if (srcPath.isEmpty() || dstPath.isEmpty()) {
return;
}
InputStream is = null;
OutputStream os = null;
try {
is = new BufferedInputStream(appCtx.getAssets().open(srcPath));
os = new BufferedOutputStream(new FileOutputStream(new File(dstPath)));
byte[] buffer = new byte[1024];
int length = 0;
while ((length = is.read(buffer)) != -1) {
os.write(buffer, 0, length);
}
} catch (FileNotFoundException e) {
e.printStackTrace();
} catch (IOException e) {
e.printStackTrace();
} finally {
try {
os.close();
is.close();
} catch (IOException e) {
e.printStackTrace();
}
}
}
public static void copyDirectoryFromAssets(Context appCtx, String srcDir, String dstDir) {
if (srcDir.isEmpty() || dstDir.isEmpty()) {
return;
}
try {
if (!new File(dstDir).exists()) {
new File(dstDir).mkdirs();
}
for (String fileName : appCtx.getAssets().list(srcDir)) {
String srcSubPath = srcDir + File.separator + fileName;
String dstSubPath = dstDir + File.separator + fileName;
if (new File(srcSubPath).isDirectory()) {
copyDirectoryFromAssets(appCtx, srcSubPath, dstSubPath);
} else {
copyFileFromAssets(appCtx, srcSubPath, dstSubPath);
}
}
} catch (Exception e) {
e.printStackTrace();
}
}
public static float[] parseFloatsFromString(String string, String delimiter) {
String[] pieces = string.trim().toLowerCase().split(delimiter);
float[] floats = new float[pieces.length];
for (int i = 0; i < pieces.length; i++) {
floats[i] = Float.parseFloat(pieces[i].trim());
}
return floats;
}
public static long[] parseLongsFromString(String string, String delimiter) {
String[] pieces = string.trim().toLowerCase().split(delimiter);
long[] longs = new long[pieces.length];
for (int i = 0; i < pieces.length; i++) {
longs[i] = Long.parseLong(pieces[i].trim());
}
return longs;
}
public static String getSDCardDirectory() {
return Environment.getExternalStorageDirectory().getAbsolutePath();
}
public static boolean isSupportedNPU() {
return false;
// String hardware = android.os.Build.HARDWARE;
// return hardware.equalsIgnoreCase("kirin810") || hardware.equalsIgnoreCase("kirin990");
}
public static Bitmap resizeWithStep(Bitmap bitmap, int maxLength, int step) {
int width = bitmap.getWidth();
int height = bitmap.getHeight();
int maxWH = Math.max(width, height);
float ratio = 1;
int newWidth = width;
int newHeight = height;
if (maxWH > maxLength) {
ratio = maxLength * 1.0f / maxWH;
newWidth = (int) Math.floor(ratio * width);
newHeight = (int) Math.floor(ratio * height);
}
newWidth = newWidth - newWidth % step;
if (newWidth == 0) {
newWidth = step;
}
newHeight = newHeight - newHeight % step;
if (newHeight == 0) {
newHeight = step;
}
return Bitmap.createScaledBitmap(bitmap, newWidth, newHeight, true);
}
}
deploy/android_demo/app/src/main/res/drawable-v24/ic_launcher_foreground.xml 0 → 100644
浏览文件 @ 26219d5f
<vector xmlns:android="http://schemas.android.com/apk/res/android"
xmlns:aapt="http://schemas.android.com/aapt"
android:width="108dp"
android:height="108dp"
android:viewportWidth="108"
android:viewportHeight="108">
<path
android:fillType="evenOdd"
android:pathData="M32,64C32,64 38.39,52.99 44.13,50.95C51.37,48.37 70.14,49.57 70.14,49.57L108.26,87.69L108,109.01L75.97,107.97L32,64Z"
android:strokeWidth="1"
android:strokeColor="#00000000">
<aapt:attr name="android:fillColor">
<gradient
android:endX="78.5885"
android:endY="90.9159"
android:startX="48.7653"
android:startY="61.0927"
android:type="linear">
<item
android:color="#44000000"
android:offset="0.0" />
<item
android:color="#00000000"
android:offset="1.0" />
</gradient>
</aapt:attr>
</path>
<path
android:fillColor="#FFFFFF"
android:fillType="nonZero"
android:pathData="M66.94,46.02L66.94,46.02C72.44,50.07 76,56.61 76,64L32,64C32,56.61 35.56,50.11 40.98,46.06L36.18,41.19C35.45,40.45 35.45,39.3 36.18,38.56C36.91,37.81 38.05,37.81 38.78,38.56L44.25,44.05C47.18,42.57 50.48,41.71 54,41.71C57.48,41.71 60.78,42.57 63.68,44.05L69.11,38.56C69.84,37.81 70.98,37.81 71.71,38.56C72.44,39.3 72.44,40.45 71.71,41.19L66.94,46.02ZM62.94,56.92C64.08,56.92 65,56.01 65,54.88C65,53.76 64.08,52.85 62.94,52.85C61.8,52.85 60.88,53.76 60.88,54.88C60.88,56.01 61.8,56.92 62.94,56.92ZM45.06,56.92C46.2,56.92 47.13,56.01 47.13,54.88C47.13,53.76 46.2,52.85 45.06,52.85C43.92,52.85 43,53.76 43,54.88C43,56.01 43.92,56.92 45.06,56.92Z"
android:strokeWidth="1"
android:strokeColor="#00000000" />
</vector>
deploy/android_demo/app/src/main/res/drawable/ic_launcher_background.xml 0 → 100644
浏览文件 @ 26219d5f
<?xml version="1.0" encoding="utf-8"?>
<vector xmlns:android="http://schemas.android.com/apk/res/android"
android:width="108dp"
android:height="108dp"
android:viewportWidth="108"
android:viewportHeight="108">
<path
android:fillColor="#008577"
android:pathData="M0,0h108v108h-108z" />
<path
android:fillColor="#00000000"
android:pathData="M9,0L9,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M19,0L19,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M29,0L29,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M39,0L39,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M49,0L49,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M59,0L59,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M69,0L69,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M79,0L79,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M89,0L89,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M99,0L99,108"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,9L108,9"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,19L108,19"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,29L108,29"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,39L108,39"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,49L108,49"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,59L108,59"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,69L108,69"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,79L108,79"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,89L108,89"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M0,99L108,99"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M19,29L89,29"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M19,39L89,39"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M19,49L89,49"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M19,59L89,59"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M19,69L89,69"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M19,79L89,79"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M29,19L29,89"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M39,19L39,89"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M49,19L49,89"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M59,19L59,89"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M69,19L69,89"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
<path
android:fillColor="#00000000"
android:pathData="M79,19L79,89"
android:strokeWidth="0.8"
android:strokeColor="#33FFFFFF" />
</vector>
deploy/android_demo/app/src/main/res/layout/activity_main.xml 0 → 100644
浏览文件 @ 26219d5f
<?xml version="1.0" encoding="utf-8"?>
<android.support.constraint.ConstraintLayout xmlns:android="http://schemas.android.com/apk/res/android"
xmlns:app="http://schemas.android.com/apk/res-auto"
xmlns:tools="http://schemas.android.com/tools"
android:layout_width="match_parent"
android:layout_height="match_parent"
tools:context=".MainActivity">
<RelativeLayout
android:layout_width="match_parent"
android:layout_height="match_parent">
<LinearLayout
android:id="@+id/v_input_info"
android:layout_width="fill_parent"
android:layout_height="wrap_content"
android:layout_alignParentTop="true"
android:orientation="vertical">
<TextView
android:id="@+id/tv_input_setting"
android:layout_width="wrap_content"
android:layout_height="wrap_content"
android:scrollbars="vertical"
android:layout_marginLeft="12dp"
android:layout_marginRight="12dp"
android:layout_marginTop="10dp"
android:layout_marginBottom="5dp"
android:lineSpacingExtra="4dp"
android:singleLine="false"
android:maxLines="6"
android:text=""/>
</LinearLayout>
<RelativeLayout
android:layout_width="match_parent"
android:layout_height="match_parent"
android:layout_above="@+id/v_output_info"
android:layout_below="@+id/v_input_info">
<ImageView
android:id="@+id/iv_input_image"
android:layout_width="400dp"
android:layout_height="400dp"
android:layout_centerHorizontal="true"
android:layout_centerVertical="true"
android:layout_marginLeft="12dp"
android:layout_marginRight="12dp"
android:layout_marginTop="5dp"
android:layout_marginBottom="5dp"
android:adjustViewBounds="true"
android:scaleType="fitCenter"/>
</RelativeLayout>
<RelativeLayout
android:id="@+id/v_output_info"
android:layout_width="wrap_content"
android:layout_height="wrap_content"
android:layout_alignParentBottom="true"
android:layout_centerHorizontal="true">
<TextView
android:id="@+id/tv_output_result"
android:layout_width="wrap_content"
android:layout_height="wrap_content"
android:layout_alignParentTop="true"
android:layout_centerHorizontal="true"
android:layout_centerVertical="true"
android:scrollbars="vertical"
android:layout_marginLeft="12dp"
android:layout_marginRight="12dp"
android:layout_marginTop="5dp"
android:layout_marginBottom="5dp"
android:textAlignment="center"
android:lineSpacingExtra="5dp"
android:singleLine="false"
android:maxLines="5"
android:text=""/>
<TextView
android:id="@+id/tv_inference_time"
android:layout_width="wrap_content"
android:layout_height="wrap_content"
android:layout_below="@+id/tv_output_result"
android:layout_centerHorizontal="true"
android:layout_centerVertical="true"
android:textAlignment="center"
android:layout_marginLeft="12dp"
android:layout_marginRight="12dp"
android:layout_marginTop="5dp"
android:layout_marginBottom="10dp"
android:text=""/>
</RelativeLayout>
</RelativeLayout>
</android.support.constraint.ConstraintLayout>
\ No newline at end of file
deploy/android_demo/app/src/main/res/menu/menu_action_options.xml 0 → 100644
浏览文件 @ 26219d5f
<menu xmlns:android="http://schemas.android.com/apk/res/android"
xmlns:app="http://schemas.android.com/apk/res-auto">
<group android:id="@+id/pick_image">
<item
android:id="@+id/open_gallery"
android:title="Open Gallery"
app:showAsAction="withText"/>
<item
android:id="@+id/take_photo"
android:title="Take Photo"
app:showAsAction="withText"/>
</group>
<group>
<item
android:id="@+id/settings"
android:title="Settings..."
app:showAsAction="withText"/>
</group>
</menu>
deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher.xml 0 → 100644
浏览文件 @ 26219d5f
<?xml version="1.0" encoding="utf-8"?>
<adaptive-icon xmlns:android="http://schemas.android.com/apk/res/android">
<background android:drawable="@drawable/ic_launcher_background" />
<foreground android:drawable="@drawable/ic_launcher_foreground" />
</adaptive-icon>
\ No newline at end of file
deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher_round.xml 0 → 100644
浏览文件 @ 26219d5f
<?xml version="1.0" encoding="utf-8"?>
<adaptive-icon xmlns:android="http://schemas.android.com/apk/res/android">
<background android:drawable="@drawable/ic_launcher_background" />
<foreground android:drawable="@drawable/ic_launcher_foreground" />
</adaptive-icon>
\ No newline at end of file
deploy/android_demo/app/src/main/res/values/arrays.xml 0 → 100644
浏览文件 @ 26219d5f
<?xml version="1.0" encoding="utf-8"?>
<resources>
<string-array name="cpu_thread_num_entries">
<item>1 threads</item>
<item>2 threads</item>
<item>4 threads</item>
<item>8 threads</item>
</string-array>
<string-array name="cpu_thread_num_values">
<item>1</item>
<item>2</item>
<item>4</item>
<item>8</item>
</string-array>
<string-array name="cpu_power_mode_entries">
<item>HIGH(only big cores)</item>
<item>LOW(only LITTLE cores)</item>
<item>FULL(all cores)</item>
<item>NO_BIND(depends on system)</item>
<item>RAND_HIGH</item>
<item>RAND_LOW</item>
</string-array>
<string-array name="cpu_power_mode_values">
<item>LITE_POWER_HIGH</item>
<item>LITE_POWER_LOW</item>
<item>LITE_POWER_FULL</item>
<item>LITE_POWER_NO_BIND</item>
<item>LITE_POWER_RAND_HIGH</item>
<item>LITE_POWER_RAND_LOW</item>
</string-array>
<string-array name="input_color_format_entries">
<item>BGR color format</item>
<item>RGB color format</item>
</string-array>
<string-array name="input_color_format_values">
<item>BGR</item>
<item>RGB</item>
</string-array>
</resources>
\ No newline at end of file
deploy/android_demo/app/src/main/res/values/colors.xml 0 → 100644
浏览文件 @ 26219d5f
<?xml version="1.0" encoding="utf-8"?>
<resources>
<color name="colorPrimary">#008577</color>
<color name="colorPrimaryDark">#00574B</color>
<color name="colorAccent">#D81B60</color>
</resources>
deploy/android_demo/app/src/main/res/values/strings.xml 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/android_demo/app/src/main/res/values/styles.xml 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/android_demo/app/src/main/res/xml/settings.xml 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/android_demo/app/src/test/java/com/baidu/paddle/lite/demo/ocr/ExampleUnitTest.java 0 → 100644
浏览文件 @ 26219d5f
package com.baidu.paddle.lite.demo.ocr;
import org.junit.Test;
import static org.junit.Assert.*;
/**
* Example local unit test, which will execute on the development machine (host).
*
* @see <a href="http://d.android.com/tools/testing">Testing documentation</a>
*/
public class ExampleUnitTest {
@Test
public void addition_isCorrect() {
assertEquals(4, 2 + 2);
}
}
\ No newline at end of file
deploy/android_demo/build.gradle 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/android_demo/gradle.properties 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/android_demo/gradlew 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/android_demo/gradlew.bat 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/CMakeLists.txt 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/include/clipper.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/include/config.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/include/ocr_det.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/include/ocr_rec.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/include/postprocess_op.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/include/preprocess_op.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/include/utility.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/readme.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/clipper.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/config.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/main.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/ocr_det.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/ocr_rec.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/postprocess_op.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/preprocess_op.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/src/utility.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/tools/build.sh 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/tools/config.txt 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/cpp_infer/tools/run.sh 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_det/config.json 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_det/module.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_det/params.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_rec/config.json 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_rec/module.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_rec/params.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_system/config.json 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_system/module.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/hubserving/ocr_system/params.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/imgs/cpp_infer_pred_12.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/imgs/demo.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/download_dependencies.sh 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo.xcodeproj/project.pbxproj 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/AppDelegate.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/AppDelegate.m 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/Assets.xcassets/Contents.json 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/Base.lproj/Main.storyboard 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/BoxLayer.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/BoxLayer.m 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/Helpers.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/Helpers.m 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/Info.plist 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/OcrData.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/OcrData.m 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/ViewController.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/ViewController.mm 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/label_list.txt 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/main.m 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/ocr.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/pdocr/ocr_clipper.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/pdocr/ocr_clipper.hpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/pdocr/ocr_crnn_process.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/pdocr/ocr_crnn_process.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/pdocr/ocr_db_post_process.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/pdocr/ocr_db_post_process.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/ios_demo/ocr_demo/timer.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/Makefile 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/config.txt 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/crnn_process.cc 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/crnn_process.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/db_post_process.cc 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/db_post_process.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/ocr_db_crnn.cc 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
deploy/lite/readme.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/ArT.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/CASIA_0.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/LSVT_1.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/LSVT_2.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/captcha_demo.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/ccpd_demo.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/ch_doc1.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/ch_doc2.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/ch_doc3.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/ch_street_rec_1.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/ch_street_rec_2.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/cmb_demo.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/labelimg.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/labelme.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/nist_demo.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/rctw.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/datasets/roLabelImg.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/FAQ.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/benchmark.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/customize.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/data_annotation.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/data_synthesis.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/datasets.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/handwritten_datasets.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/quickstart.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/reference.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/serving.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/tricks.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/update.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/vertical_and_multilingual_datasets.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_ch/visualization.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/FAQ_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/config_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/customize_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/datasets_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/detection_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/inference_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/installation_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/recognition_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/doc_en/update_en.md 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_en/img_12.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/1.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/10.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/11.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/12.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/13.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/15.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/16.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/17.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/2.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/22.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/3.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/4.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/5.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/6.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/7.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/8.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/9.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/chinese_db_crnn_server/en_paper.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results/img_11.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/1.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/10.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/11.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/12.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/13.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/15.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/16.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/17.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/2.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/22.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/3.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/4.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/5.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/6.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/7.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/8.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/imgs_results_vis2/9.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/joinus.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/ocr-android-easyedge.png 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
doc/tricks/long_text_examples.jpg 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/data/det/dataset_traversal.py 100755 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/data/det/db_process.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/data/det/east_process.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/data/reader_main.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/data/rec/dataset_traversal.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/data/rec/img_tools.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/modeling/architectures/__init__.py 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/modeling/backbones/__init__.py 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/modeling/heads/__init__.py 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/modeling/heads/det_db_head.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/modeling/losses/__init__.py 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/modeling/stns/__init__.py 0 → 100755
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/optimizer.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/db_postprocess.py
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/.gitignore 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/.ycm_extra_conf.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/Makefile 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/__init__.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/__main__.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/adaptor.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/clipper/clipper.cpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/clipper/clipper.hpp 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/attr.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/buffer_info.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/cast.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/chrono.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/common.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/complex.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/descr.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/eigen.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/embed.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/eval.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/functional.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/numpy.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/operators.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/options.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/pybind11.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/pytypes.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/stl.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/stl_bind.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/include/pybind11/typeid.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/postprocess/lanms/lanms.h 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
ppocr/utils/character.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/eval.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/eval_utils/eval_det_utils.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/eval_utils/eval_rec_utils.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/export_model.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/infer/predict_det.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/infer/predict_rec.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/infer/predict_system.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/infer/utility.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/infer_det.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/infer_rec.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/program.py
浏览文件 @ 26219d5f
此差异已折叠。
tools/test_hubserving.py 0 → 100644
浏览文件 @ 26219d5f
此差异已折叠。
tools/train.py
浏览文件 @ 26219d5f
此差异已折叠。
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或