提交 d25e263a 编写于 作者: A andyjpaddle

Merge branch 'dygraph' of https://github.com/PaddlePaddle/PaddleOCR into dygraph

...@@ -101,7 +101,7 @@ PaddleOCR support a variety of cutting-edge algorithms related to OCR, and devel ...@@ -101,7 +101,7 @@ PaddleOCR support a variety of cutting-edge algorithms related to OCR, and devel
- [PP-Structure 🔥](./ppstructure/README.md) - [PP-Structure 🔥](./ppstructure/README.md)
- [Quick Start](./ppstructure/docs/quickstart_en.md) - [Quick Start](./ppstructure/docs/quickstart_en.md)
- [Model Zoo](./ppstructure/docs/models_list_en.md) - [Model Zoo](./ppstructure/docs/models_list_en.md)
- [Model training](./doc/doc_en/training_en.md) - [Model training](./doc/doc_en/training_en.md)
- [Layout Parser](./ppstructure/layout/README.md) - [Layout Parser](./ppstructure/layout/README.md)
- [Table Recognition](./ppstructure/table/README.md) - [Table Recognition](./ppstructure/table/README.md)
- [DocVQA](./ppstructure/vqa/README.md) - [DocVQA](./ppstructure/vqa/README.md)
...@@ -121,9 +121,9 @@ PaddleOCR support a variety of cutting-edge algorithms related to OCR, and devel ...@@ -121,9 +121,9 @@ PaddleOCR support a variety of cutting-edge algorithms related to OCR, and devel
- [Other Data Annotation Tools](./doc/doc_en/data_annotation_en.md) - [Other Data Annotation Tools](./doc/doc_en/data_annotation_en.md)
- [Other Data Synthesis Tools](./doc/doc_en/data_synthesis_en.md) - [Other Data Synthesis Tools](./doc/doc_en/data_synthesis_en.md)
- Datasets - Datasets
- [General OCR Datasets(Chinese/English)](./doc/doc_en/datasets_en.md) - [General OCR Datasets(Chinese/English)](doc/doc_en/dataset/datasets_en.md)
- [HandWritten_OCR_Datasets(Chinese)](./doc/doc_en/handwritten_datasets_en.md) - [HandWritten_OCR_Datasets(Chinese)](doc/doc_en/dataset/handwritten_datasets_en.md)
- [Various OCR Datasets(multilingual)](./doc/doc_en/vertical_and_multilingual_datasets_en.md) - [Various OCR Datasets(multilingual)](doc/doc_en/dataset/vertical_and_multilingual_datasets_en.md)
- [Code Structure](./doc/doc_en/tree_en.md) - [Code Structure](./doc/doc_en/tree_en.md)
- [Visualization](#Visualization) - [Visualization](#Visualization)
- [Community](#Community) - [Community](#Community)
...@@ -170,4 +170,4 @@ More details, please refer to [Multilingual OCR Development Plan](https://github ...@@ -170,4 +170,4 @@ More details, please refer to [Multilingual OCR Development Plan](https://github
<a name="LICENSE"></a> <a name="LICENSE"></a>
## License ## License
This project is released under <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/master/LICENSE">Apache 2.0 license</a> This project is released under <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/master/LICENSE">Apache 2.0 license</a>
\ No newline at end of file
...@@ -128,12 +128,12 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力 ...@@ -128,12 +128,12 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力
- [其它数据标注工具](./doc/doc_ch/data_annotation.md) - [其它数据标注工具](./doc/doc_ch/data_annotation.md)
- [其它数据合成工具](./doc/doc_ch/data_synthesis.md) - [其它数据合成工具](./doc/doc_ch/data_synthesis.md)
- 数据集 - 数据集
- [通用中英文OCR数据集](./doc/doc_ch/datasets.md) - [通用中英文OCR数据集](doc/doc_ch/dataset/datasets.md)
- [手写中文OCR数据集](./doc/doc_ch/handwritten_datasets.md) - [手写中文OCR数据集](doc/doc_ch/dataset/handwritten_datasets.md)
- [垂类多语言OCR数据集](./doc/doc_ch/vertical_and_multilingual_datasets.md) - [垂类多语言OCR数据集](doc/doc_ch/dataset/vertical_and_multilingual_datasets.md)
- [版面分析数据集](./doc/doc_ch/layout_datasets.md) - [版面分析数据集](doc/doc_ch/dataset/layout_datasets.md)
- [表格识别数据集](./doc/doc_ch/table_datasets.md) - [表格识别数据集](doc/doc_ch/dataset/table_datasets.md)
- [DocVQA数据集](./doc/doc_ch/docvqa_datasets.md) - [DocVQA数据集](doc/doc_ch/dataset/docvqa_datasets.md)
- [代码组织结构](./doc/doc_ch/tree.md) - [代码组织结构](./doc/doc_ch/tree.md)
- [效果展示](#效果展示) - [效果展示](#效果展示)
- [《动手学OCR》电子书📚](./doc/doc_ch/ocr_book.md) - [《动手学OCR》电子书📚](./doc/doc_ch/ocr_book.md)
...@@ -160,13 +160,13 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力 ...@@ -160,13 +160,13 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力
<img src="./doc/imgs_results/ch_ppocr_mobile_v2.0/00056221.jpg" width="800"> <img src="./doc/imgs_results/ch_ppocr_mobile_v2.0/00056221.jpg" width="800">
<img src="./doc/imgs_results/ch_ppocr_mobile_v2.0/rotate_00052204.jpg" width="800"> <img src="./doc/imgs_results/ch_ppocr_mobile_v2.0/rotate_00052204.jpg" width="800">
</div> </div>
</details> </details>
<details open> <details open>
<summary>PP-OCRv2 英文模型</summary> <summary>PP-OCRv2 英文模型</summary>
<div align="center"> <div align="center">
<img src="./doc/imgs_results/ch_ppocr_mobile_v2.0/img_12.jpg" width="800"> <img src="./doc/imgs_results/ch_ppocr_mobile_v2.0/img_12.jpg" width="800">
</div> </div>
...@@ -176,12 +176,12 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力 ...@@ -176,12 +176,12 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力
<details open> <details open>
<summary>PP-OCRv2 其他语言模型</summary> <summary>PP-OCRv2 其他语言模型</summary>
<div align="center"> <div align="center">
<img src="./doc/imgs_results/french_0.jpg" width="800"> <img src="./doc/imgs_results/french_0.jpg" width="800">
<img src="./doc/imgs_results/korean.jpg" width="800"> <img src="./doc/imgs_results/korean.jpg" width="800">
</div> </div>
</details> </details>
<details open> <details open>
...@@ -196,8 +196,8 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力 ...@@ -196,8 +196,8 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库,助力
<div align="center"> <div align="center">
<img src="./ppstructure/docs/vqa/result_ser/zh_val_0_ser.jpg" width="800"> <img src="./ppstructure/docs/vqa/result_ser/zh_val_0_ser.jpg" width="800">
</div> </div>
- RE(关系提取) - RE(关系提取)
<div align="center"> <div align="center">
<img src="./ppstructure/docs/vqa/result_re/zh_val_21_re.jpg" width="800"> <img src="./ppstructure/docs/vqa/result_re/zh_val_21_re.jpg" width="800">
</div> </div>
......
Global:
debug: false
use_gpu: true
epoch_num: 500
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/ch_PP-OCR_v3_det/
save_epoch_step: 100
eval_batch_step:
- 0
- 400
cal_metric_during_train: false
pretrained_model: null
checkpoints: null
save_inference_dir: null
use_visualdl: false
infer_img: doc/imgs_en/img_10.jpg
save_res_path: ./checkpoints/det_db/predicts_db.txt
distributed: true
Architecture:
name: DistillationModel
algorithm: Distillation
model_type: det
Models:
Student:
model_type: det
algorithm: DB
Transform: null
Backbone:
name: MobileNetV3
scale: 0.5
model_name: large
disable_se: true
Neck:
name: RSEFPN
out_channels: 96
shortcut: True
Head:
name: DBHead
k: 50
Student2:
model_type: det
algorithm: DB
Transform: null
Backbone:
name: MobileNetV3
scale: 0.5
model_name: large
disable_se: true
Neck:
name: RSEFPN
out_channels: 96
shortcut: True
Head:
name: DBHead
k: 50
Teacher:
freeze_params: true
return_all_feats: false
model_type: det
algorithm: DB
Backbone:
name: ResNet
in_channels: 3
layers: 50
Neck:
name: LKPAN
out_channels: 256
Head:
name: DBHead
kernel_list: [7,2,2]
k: 50
Loss:
name: CombinedLoss
loss_config_list:
- DistillationDilaDBLoss:
weight: 1.0
model_name_pairs:
- ["Student", "Teacher"]
- ["Student2", "Teacher"]
key: maps
balance_loss: true
main_loss_type: DiceLoss
alpha: 5
beta: 10
ohem_ratio: 3
- DistillationDMLLoss:
model_name_pairs:
- ["Student", "Student2"]
maps_name: "thrink_maps"
weight: 1.0
# act: None
model_name_pairs: ["Student", "Student2"]
key: maps
- DistillationDBLoss:
weight: 1.0
model_name_list: ["Student", "Student2"]
# key: maps
# name: DBLoss
balance_loss: true
main_loss_type: DiceLoss
alpha: 5
beta: 10
ohem_ratio: 3
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Cosine
learning_rate: 0.001
warmup_epoch: 2
regularizer:
name: L2
factor: 5.0e-05
PostProcess:
name: DistillationDBPostProcess
model_name: ["Student"]
key: head_out
thresh: 0.3
box_thresh: 0.6
max_candidates: 1000
unclip_ratio: 1.5
Metric:
name: DistillationMetric
base_metric_name: DetMetric
main_indicator: hmean
key: "Student"
Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
ratio_list: [1.0]
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- DetLabelEncode: null
- CopyPaste:
- IaaAugment:
augmenter_args:
- type: Fliplr
args:
p: 0.5
- type: Affine
args:
rotate:
- -10
- 10
- type: Resize
args:
size:
- 0.5
- 3
- EastRandomCropData:
size:
- 960
- 960
max_tries: 50
keep_ratio: true
- MakeBorderMap:
shrink_ratio: 0.4
thresh_min: 0.3
thresh_max: 0.7
- MakeShrinkMap:
shrink_ratio: 0.4
min_text_size: 8
- NormalizeImage:
scale: 1./255.
mean:
- 0.485
- 0.456
- 0.406
std:
- 0.229
- 0.224
- 0.225
order: hwc
- ToCHWImage: null
- KeepKeys:
keep_keys:
- image
- threshold_map
- threshold_mask
- shrink_map
- shrink_mask
loader:
shuffle: true
drop_last: false
batch_size_per_card: 8
num_workers: 4
Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- DetLabelEncode: null
- DetResizeForTest: null
- NormalizeImage:
scale: 1./255.
mean:
- 0.485
- 0.456
- 0.406
std:
- 0.229
- 0.224
- 0.225
order: hwc
- ToCHWImage: null
- KeepKeys:
keep_keys:
- image
- shape
- polys
- ignore_tags
loader:
shuffle: false
drop_last: false
batch_size_per_card: 1
num_workers: 2
Global:
debug: false
use_gpu: true
epoch_num: 500
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/ch_PP-OCR_V3_det/
save_epoch_step: 100
eval_batch_step:
- 0
- 400
cal_metric_during_train: false
pretrained_model: null
checkpoints: null
save_inference_dir: null
use_visualdl: false
infer_img: doc/imgs_en/img_10.jpg
save_res_path: ./checkpoints/det_db/predicts_db.txt
distributed: true
Architecture:
model_type: det
algorithm: DB
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: large
disable_se: True
Neck:
name: RSEFPN
out_channels: 96
shortcut: True
Head:
name: DBHead
k: 50
Loss:
name: DBLoss
balance_loss: true
main_loss_type: DiceLoss
alpha: 5
beta: 10
ohem_ratio: 3
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Cosine
learning_rate: 0.001
warmup_epoch: 2
regularizer:
name: L2
factor: 5.0e-05
PostProcess:
name: DBPostProcess
thresh: 0.3
box_thresh: 0.6
max_candidates: 1000
unclip_ratio: 1.5
Metric:
name: DetMetric
main_indicator: hmean
Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
ratio_list: [1.0]
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- DetLabelEncode: null
- IaaAugment:
augmenter_args:
- type: Fliplr
args:
p: 0.5
- type: Affine
args:
rotate:
- -10
- 10
- type: Resize
args:
size:
- 0.5
- 3
- EastRandomCropData:
size:
- 960
- 960
max_tries: 50
keep_ratio: true
- MakeBorderMap:
shrink_ratio: 0.4
thresh_min: 0.3
thresh_max: 0.7
- MakeShrinkMap:
shrink_ratio: 0.4
min_text_size: 8
- NormalizeImage:
scale: 1./255.
mean:
- 0.485
- 0.456
- 0.406
std:
- 0.229
- 0.224
- 0.225
order: hwc
- ToCHWImage: null
- KeepKeys:
keep_keys:
- image
- threshold_map
- threshold_mask
- shrink_map
- shrink_mask
loader:
shuffle: true
drop_last: false
batch_size_per_card: 8
num_workers: 4
Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data/icdar2015/text_localization/
label_file_list:
- ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- DetLabelEncode: null
- DetResizeForTest: null
- NormalizeImage:
scale: 1./255.
mean:
- 0.485
- 0.456
- 0.406
std:
- 0.229
- 0.224
- 0.225
order: hwc
- ToCHWImage: null
- KeepKeys:
keep_keys:
- image
- shape
- polys
- ignore_tags
loader:
shuffle: false
drop_last: false
batch_size_per_card: 1
num_workers: 2
# FCENet
- [1. 算法简介](#1)
- [2. 环境配置](#2)
- [3. 模型训练、评估、预测](#3)
- [3.1 训练](#3-1)
- [3.2 评估](#3-2)
- [3.3 预测](#3-3)
- [4. 推理部署](#4)
- [4.1 Python推理](#4-1)
- [4.2 C++推理](#4-2)
- [4.3 Serving服务化部署](#4-3)
- [4.4 更多推理部署](#4-4)
- [5. FAQ](#5)
<a name="1"></a>
## 1. 算法简介
论文信息:
> [Fourier Contour Embedding for Arbitrary-Shaped Text Detection](https://arxiv.org/abs/2104.10442)
> Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang
> CVPR, 2021
在CTW1500文本检测公开数据集上,算法复现效果如下:
| 模型 |骨干网络|配置文件|precision|recall|Hmean|下载链接|
|-----| --- | --- | --- | --- | --- | --- |
| FCE | ResNet50_dcn | [configs/det/det_r50_vd_dcn_fce_ctw.yml](../../configs/det/det_r50_vd_dcn_fce_ctw.yml)| 88.39%|82.18%|85.27%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar)|
<a name="2"></a>
## 2. 环境配置
请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境,参考[《项目克隆》](./clone.md)克隆项目代码。
<a name="3"></a>
## 3. 模型训练、评估、预测
上述FCE模型使用CTW1500文本检测公开数据集训练得到,数据集下载可参考 [ocr_datasets](./dataset/ocr_datasets.md)
数据下载完成后,请参考[文本检测训练教程](./detection.md)进行训练。PaddleOCR对代码进行了模块化,训练不同的检测模型只需要**更换配置文件**即可。
<a name="4"></a>
## 4. 推理部署
<a name="4-1"></a>
### 4.1 Python推理
首先将FCE文本检测训练过程中保存的模型,转换成inference model。以基于Resnet50_vd_dcn骨干网络,在CTW1500英文数据集训练的模型为例( [模型下载地址](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar) ),可以使用如下命令进行转换:
```shell
python3 tools/export_model.py -c configs/det/det_r50_vd_dcn_fce_ctw.yml -o Global.pretrained_model=./det_r50_dcn_fce_ctw_v2.0_train/best_accuracy Global.save_inference_dir=./inference/det_fce
```
FCE文本检测模型推理,执行非弯曲文本检测,可以执行如下命令:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=quad
```
可视化文本检测结果默认保存到`./inference_results`文件夹里面,结果文件的名称前缀为'det_res'。结果示例如下:
![](../imgs_results/det_res_img_10_fce.jpg)
如果想执行弯曲文本检测,可以执行如下命令:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=poly
```
可视化文本检测结果默认保存到`./inference_results`文件夹里面,结果文件的名称前缀为'det_res'。结果示例如下:
![](../imgs_results/det_res_img623_fce.jpg)
**注意**:由于CTW1500数据集只有1000张训练图像,且主要针对英文场景,所以上述模型对中文文本图像检测效果会比较差。
<a name="4-2"></a>
### 4.2 C++推理
由于后处理暂未使用CPP编写,FCE文本检测模型暂不支持CPP推理。
<a name="4-3"></a>
### 4.3 Serving服务化部署
暂未支持
<a name="4-4"></a>
### 4.4 更多推理部署
暂未支持
<a name="5"></a>
## 5. FAQ
## 引用
```bibtex
@InProceedings{zhu2021fourier,
title={Fourier Contour Embedding for Arbitrary-Shaped Text Detection},
author={Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang},
year={2021},
booktitle = {CVPR}
}
```
# PSENet
- [1. 算法简介](#1)
- [2. 环境配置](#2)
- [3. 模型训练、评估、预测](#3)
- [3.1 训练](#3-1)
- [3.2 评估](#3-2)
- [3.3 预测](#3-3)
- [4. 推理部署](#4)
- [4.1 Python推理](#4-1)
- [4.2 C++推理](#4-2)
- [4.3 Serving服务化部署](#4-3)
- [4.4 更多推理部署](#4-4)
- [5. FAQ](#5)
<a name="1"></a>
## 1. 算法简介
论文信息:
> [Shape robust text detection with progressive scale expansion network](https://arxiv.org/abs/1903.12473)
> Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai
> CVPR, 2019
在ICDAR2015文本检测公开数据集上,算法复现效果如下:
|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
| --- | --- | --- | --- | --- | --- | --- |
|PSE| ResNet50_vd | [configs/det/det_r50_vd_pse.yml](../../configs/det/det_r50_vd_pse.yml)| 85.81% |79.53%|82.55%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
|PSE| MobileNetV3| [configs/det/det_mv3_pse.yml](../../configs/det/det_mv3_pse.yml) | 82.20% |70.48%|75.89%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
<a name="2"></a>
## 2. 环境配置
请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境,参考[《项目克隆》](./clone.md)克隆项目代码。
<a name="3"></a>
## 3. 模型训练、评估、预测
上述PSE模型使用ICDAR2015文本检测公开数据集训练得到,数据集下载可参考 [ocr_datasets](./dataset/ocr_datasets.md)
数据下载完成后,请参考[文本检测训练教程](./detection.md)进行训练。PaddleOCR对代码进行了模块化,训练不同的检测模型只需要**更换配置文件**即可。
<a name="4"></a>
## 4. 推理部署
<a name="4-1"></a>
### 4.1 Python推理
首先将PSE文本检测训练过程中保存的模型,转换成inference model。以基于Resnet50_vd骨干网络,在ICDAR2015英文数据集训练的模型为例( [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar) ),可以使用如下命令进行转换:
```shell
python3 tools/export_model.py -c configs/det/det_r50_vd_pse.yml -o Global.pretrained_model=./det_r50_vd_pse_v2.0_train/best_accuracy Global.save_inference_dir=./inference/det_pse
```
PSE文本检测模型推理,执行非弯曲文本检测,可以执行如下命令:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=quad
```
可视化文本检测结果默认保存到`./inference_results`文件夹里面,结果文件的名称前缀为'det_res'。结果示例如下:
![](../imgs_results/det_res_img_10_pse.jpg)
如果想执行弯曲文本检测,可以执行如下命令:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=poly
```
可视化文本检测结果默认保存到`./inference_results`文件夹里面,结果文件的名称前缀为'det_res'。结果示例如下:
![](../imgs_results/det_res_img_10_pse_poly.jpg)
**注意**:由于ICDAR2015数据集只有1000张训练图像,且主要针对英文场景,所以上述模型对中文或弯曲文本图像检测效果会比较差。
<a name="4-2"></a>
### 4.2 C++推理
由于后处理暂未使用CPP编写,PSE文本检测模型暂不支持CPP推理。
<a name="4-3"></a>
### 4.3 Serving服务化部署
暂未支持
<a name="4-4"></a>
### 4.4 更多推理部署
暂未支持
<a name="5"></a>
## 5. FAQ
## 引用
```bibtex
@inproceedings{wang2019shape,
title={Shape robust text detection with progressive scale expansion network},
author={Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={9336--9345},
year={2019}
}
```
...@@ -6,17 +6,17 @@ ...@@ -6,17 +6,17 @@
- [中文文档文字识别](#中文文档文字识别) - [中文文档文字识别](#中文文档文字识别)
- [ICDAR2019-ArT](#ICDAR2019-ArT) - [ICDAR2019-ArT](#ICDAR2019-ArT)
除了开源数据,用户还可使用合成工具自行合成,可参考[数据合成工具](./data_synthesis.md) 除了开源数据,用户还可使用合成工具自行合成,可参考[数据合成工具](../data_synthesis.md)
如果需要标注自己的数据,可参考[数据标注工具](./data_annotation.md) 如果需要标注自己的数据,可参考[数据标注工具](../data_annotation.md)
<a name="ICDAR2019-LSVT"></a> <a name="ICDAR2019-LSVT"></a>
#### 1、ICDAR2019-LSVT #### 1、ICDAR2019-LSVT
- **数据来源**:https://ai.baidu.com/broad/introduction?dataset=lsvt - **数据来源**:https://ai.baidu.com/broad/introduction?dataset=lsvt
- **数据简介**: 共45w中文街景图像,包含5w(2w测试+3w训练)全标注数据(文本坐标+文本内容),40w弱标注数据(仅文本内容),如下图所示: - **数据简介**: 共45w中文街景图像,包含5w(2w测试+3w训练)全标注数据(文本坐标+文本内容),40w弱标注数据(仅文本内容),如下图所示:
![](../datasets/LSVT_1.jpg) ![](../../datasets/LSVT_1.jpg)
(a) 全标注数据 (a) 全标注数据
![](../datasets/LSVT_2.jpg) ![](../../datasets/LSVT_2.jpg)
(b) 弱标注数据 (b) 弱标注数据
- **下载地址**:https://ai.baidu.com/broad/download?dataset=lsvt - **下载地址**:https://ai.baidu.com/broad/download?dataset=lsvt
- **说明**:其中,test数据集的label目前没有开源,如要评估结果,可以去官网提交:https://rrc.cvc.uab.es/?ch=16 - **说明**:其中,test数据集的label目前没有开源,如要评估结果,可以去官网提交:https://rrc.cvc.uab.es/?ch=16
...@@ -25,16 +25,16 @@ ...@@ -25,16 +25,16 @@
#### 2、ICDAR2017-RCTW-17 #### 2、ICDAR2017-RCTW-17
- **数据来源**:https://rctw.vlrlab.net/ - **数据来源**:https://rctw.vlrlab.net/
- **数据简介**:共包含12,000+图像,大部分图片是通过手机摄像头在野外采集的。有些是截图。这些图片展示了各种各样的场景,包括街景、海报、菜单、室内场景和手机应用程序的截图。 - **数据简介**:共包含12,000+图像,大部分图片是通过手机摄像头在野外采集的。有些是截图。这些图片展示了各种各样的场景,包括街景、海报、菜单、室内场景和手机应用程序的截图。
![](../datasets/rctw.jpg) ![](../../datasets/rctw.jpg)
- **下载地址**:https://rctw.vlrlab.net/dataset/ - **下载地址**:https://rctw.vlrlab.net/dataset/
<a name="中文街景文字识别"></a> <a name="中文街景文字识别"></a>
#### 3、中文街景文字识别 #### 3、中文街景文字识别
- **数据来源**:https://aistudio.baidu.com/aistudio/competition/detail/8 - **数据来源**:https://aistudio.baidu.com/aistudio/competition/detail/8
- **数据简介**:ICDAR2019-LSVT行识别任务,共包括29万张图片,其中21万张图片作为训练集(带标注),8万张作为测试集(无标注)。数据集采自中国街景,并由街景图片中的文字行区域(例如店铺标牌、地标等等)截取出来而形成。所有图像都经过一些预处理,将文字区域利用仿射变化,等比映射为一张高为48像素的图片,如图所示: - **数据简介**:ICDAR2019-LSVT行识别任务,共包括29万张图片,其中21万张图片作为训练集(带标注),8万张作为测试集(无标注)。数据集采自中国街景,并由街景图片中的文字行区域(例如店铺标牌、地标等等)截取出来而形成。所有图像都经过一些预处理,将文字区域利用仿射变化,等比映射为一张高为48像素的图片,如图所示:
![](../datasets/ch_street_rec_1.png) ![](../../datasets/ch_street_rec_1.png)
(a) 标注:魅派集成吊顶 (a) 标注:魅派集成吊顶
![](../datasets/ch_street_rec_2.png) ![](../../datasets/ch_street_rec_2.png)
(b) 标注:母婴用品连锁 (b) 标注:母婴用品连锁
- **下载地址** - **下载地址**
https://aistudio.baidu.com/aistudio/datasetdetail/8429 https://aistudio.baidu.com/aistudio/datasetdetail/8429
...@@ -48,15 +48,15 @@ https://aistudio.baidu.com/aistudio/datasetdetail/8429 ...@@ -48,15 +48,15 @@ https://aistudio.baidu.com/aistudio/datasetdetail/8429
- 包含汉字、英文字母、数字和标点共5990个字符(字符集合:https://github.com/YCG09/chinese_ocr/blob/master/train/char_std_5990.txt ) - 包含汉字、英文字母、数字和标点共5990个字符(字符集合:https://github.com/YCG09/chinese_ocr/blob/master/train/char_std_5990.txt )
- 每个样本固定10个字符,字符随机截取自语料库中的句子 - 每个样本固定10个字符,字符随机截取自语料库中的句子
- 图片分辨率统一为280x32 - 图片分辨率统一为280x32
![](../datasets/ch_doc1.jpg) ![](../../datasets/ch_doc1.jpg)
![](../datasets/ch_doc3.jpg) ![](../../datasets/ch_doc3.jpg)
- **下载地址**:https://pan.baidu.com/s/1QkI7kjah8SPHwOQ40rS1Pw (密码:lu7m) - **下载地址**:https://pan.baidu.com/s/1QkI7kjah8SPHwOQ40rS1Pw (密码:lu7m)
<a name="ICDAR2019-ArT"></a> <a name="ICDAR2019-ArT"></a>
#### 5、ICDAR2019-ArT #### 5、ICDAR2019-ArT
- **数据来源**:https://ai.baidu.com/broad/introduction?dataset=art - **数据来源**:https://ai.baidu.com/broad/introduction?dataset=art
- **数据简介**:共包含10,166张图像,训练集5603图,测试集4563图。由Total-Text、SCUT-CTW1500、Baidu Curved Scene Text (ICDAR2019-LSVT部分弯曲数据) 三部分组成,包含水平、多方向和弯曲等多种形状的文本。 - **数据简介**:共包含10,166张图像,训练集5603图,测试集4563图。由Total-Text、SCUT-CTW1500、Baidu Curved Scene Text (ICDAR2019-LSVT部分弯曲数据) 三部分组成,包含水平、多方向和弯曲等多种形状的文本。
![](../datasets/ArT.jpg) ![](../../datasets/ArT.jpg)
- **下载地址**:https://ai.baidu.com/broad/download?dataset=art - **下载地址**:https://ai.baidu.com/broad/download?dataset=art
## 参考文献 ## 参考文献
......
...@@ -9,7 +9,7 @@ ...@@ -9,7 +9,7 @@
- **数据简介** - **数据简介**
* 包含在线和离线两类手写数据,`HWDB1.0~1.2`总共有3895135个手写单字样本,分属7356类(7185个汉字和171个英文字母、数字、符号);`HWDB2.0~2.2`总共有5091页图像,分割为52230个文本行和1349414个文字。所有文字和文本样本均存为灰度图像。部分单字样本图片如下所示。 * 包含在线和离线两类手写数据,`HWDB1.0~1.2`总共有3895135个手写单字样本,分属7356类(7185个汉字和171个英文字母、数字、符号);`HWDB2.0~2.2`总共有5091页图像,分割为52230个文本行和1349414个文字。所有文字和文本样本均存为灰度图像。部分单字样本图片如下所示。
![](../datasets/CASIA_0.jpg) ![](../../datasets/CASIA_0.jpg)
- **下载地址**:http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html - **下载地址**:http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html
- **使用建议**:数据为单字,白色背景,可以大量合成文字行进行训练。白色背景可以处理成透明状态,方便添加各种背景。对于需要语义的情况,建议从真实语料出发,抽取单字组成文字行 - **使用建议**:数据为单字,白色背景,可以大量合成文字行进行训练。白色背景可以处理成透明状态,方便添加各种背景。对于需要语义的情况,建议从真实语料出发,抽取单字组成文字行
...@@ -22,7 +22,7 @@ ...@@ -22,7 +22,7 @@
- **数据简介**: NIST19数据集适用于手写文档和字符识别的模型训练,从3600位作者的手写样本表格中提取得到,总共包含81万张字符图片。其中9张图片示例如下。 - **数据简介**: NIST19数据集适用于手写文档和字符识别的模型训练,从3600位作者的手写样本表格中提取得到,总共包含81万张字符图片。其中9张图片示例如下。
![](../datasets/nist_demo.png) ![](../../datasets/nist_demo.png)
- **下载地址**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19) - **下载地址**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19)
# OCR数据集
- [1. 文本检测](#1-文本检测)
- [1.1 PaddleOCR 文字检测数据格式](#11-paddleocr-文字检测数据格式)
- [1.2 公开数据集](#12-公开数据集)
- [1.2.1 ICDAR 2015](#121-icdar-2015)
- [2. 文本识别](#2-文本识别)
- [2.1 PaddleOCR 文字识别数据格式](#21-paddleocr-文字识别数据格式)
- [2.2 公开数据集](#22-公开数据集)
- [2.1 ICDAR 2015](#21-icdar-2015)
- [3. 数据存放路径](#3-数据存放路径)
这里整理了OCR中常用的公开数据集,持续更新中,欢迎各位小伙伴贡献数据集~
## 1. 文本检测
### 1.1 PaddleOCR 文字检测数据格式
PaddleOCR 中的文本检测算法支持的标注文件格式如下,中间用"\t"分隔:
```
" 图像文件名 json.dumps编码的图像标注信息"
ch4_test_images/img_61.jpg [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]
```
json.dumps编码前的图像标注信息是包含多个字典的list,字典中的 `points` 表示文本框的四个点的坐标(x, y),从左上角的点开始顺时针排列。
`transcription` 表示当前文本框的文字,**当其内容为“###”时,表示该文本框无效,在训练时会跳过。**
如果您想在我们未提供的数据集上训练,可以按照上述形式构建标注文件。
### 1.2 公开数据集
| 数据集名称 |图片下载地址| PaddleOCR 标注下载地址 |
|---|---|---|
| ICDAR 2015 |https://rrc.cvc.uab.es/?ch=4&com=downloads| [train](https://paddleocr.bj.bcebos.com/dataset/train_icdar2015_label.txt) / [test](https://paddleocr.bj.bcebos.com/dataset/test_icdar2015_label.txt) |
| ctw1500 |https://paddleocr.bj.bcebos.com/dataset/ctw1500.zip| 图片下载地址中已包含 |
| total text |https://paddleocr.bj.bcebos.com/dataset/total_text.tar| 图片下载地址中已包含 |
#### 1.2.1 ICDAR 2015
ICDAR 2015 数据集包含1000张训练图像和500张测试图像。ICDAR 2015 数据集可以从上表中链接下载,首次下载需注册。
注册完成登陆后,下载下图中红色框标出的部分,其中, `Training Set Images`下载的内容保存在`icdar_c4_train_imgs`文件夹下,`Test Set Images` 下载的内容保存早`ch4_test_images`文件夹下
<p align="center">
<img src="../../datasets/ic15_location_download.png" align="middle" width = "700"/>
<p align="center">
将下载到的数据集解压到工作目录下,假设解压在 PaddleOCR/train_data/下。然后从上表中下载转换好的标注文件。
PaddleOCR 也提供了数据格式转换脚本,可以将官网 label 转换支持的数据格式。 数据转换工具在 `ppocr/utils/gen_label.py`, 这里以训练集为例:
```
# 将官网下载的标签文件转换为 train_icdar2015_label.txt
python gen_label.py --mode="det" --root_path="/path/to/icdar_c4_train_imgs/" \
--input_path="/path/to/ch4_training_localization_transcription_gt" \
--output_label="/path/to/train_icdar2015_label.txt"
```
解压数据集和下载标注文件后,PaddleOCR/train_data/ 有两个文件夹和两个文件,按照如下方式组织icdar2015数据集:
```
/PaddleOCR/train_data/icdar2015/text_localization/
└─ icdar_c4_train_imgs/ icdar 2015 数据集的训练数据
└─ ch4_test_images/ icdar 2015 数据集的测试数据
└─ train_icdar2015_label.txt icdar 2015 数据集的训练标注
└─ test_icdar2015_label.txt icdar 2015 数据集的测试标注
```
## 2. 文本识别
### 2.1 PaddleOCR 文字识别数据格式
PaddleOCR 中的文字识别算法支持两种数据格式:
- `lmdb` 用于训练以lmdb格式存储的数据集,使用 [lmdb_dataset.py](../../../ppocr/data/lmdb_dataset.py) 进行读取;
- `通用数据` 用于训练以文本文件存储的数据集,使用 [simple_dataset.py](../../../ppocr/data/simple_dataset.py)进行读取。
下面以通用数据集为例, 介绍如何准备数据集:
* 训练集
建议将训练图片放入同一个文件夹,并用一个txt文件(rec_gt_train.txt)记录图片路径和标签,txt文件里的内容如下:
**注意:** txt文件中默认请将图片路径和图片标签用 \t 分割,如用其他方式分割将造成训练报错。
```
" 图像文件名 图像标注信息 "
train_data/rec/train/word_001.jpg 简单可依赖
train_data/rec/train/word_002.jpg 用科技让复杂的世界更简单
...
```
最终训练集应有如下文件结构:
```
|-train_data
|-rec
|- rec_gt_train.txt
|- train
|- word_001.png
|- word_002.jpg
|- word_003.jpg
| ...
```
除上述单张图像为一行格式之外,PaddleOCR也支持对离线增广后的数据进行训练,为了防止相同样本在同一个batch中被多次采样,我们可以将相同标签对应的图片路径写在一行中,以列表的形式给出,在训练中,PaddleOCR会随机选择列表中的一张图片进行训练。对应地,标注文件的格式如下。
```
["11.jpg", "12.jpg"] 简单可依赖
["21.jpg", "22.jpg", "23.jpg"] 用科技让复杂的世界更简单
3.jpg ocr
```
上述示例标注文件中,"11.jpg"和"12.jpg"的标签相同,都是`简单可依赖`,在训练的时候,对于该行标注,会随机选择其中的一张图片进行训练。
- 验证集
同训练集类似,验证集也需要提供一个包含所有图片的文件夹(test)和一个rec_gt_test.txt,验证集的结构如下所示:
```
|-train_data
|-rec
|- rec_gt_test.txt
|- test
|- word_001.jpg
|- word_002.jpg
|- word_003.jpg
| ...
```
### 2.2 公开数据集
| 数据集名称 | 图片下载地址 | PaddleOCR 标注下载地址 |
|---|---|---------------------------------------------------------------------|
| en benchmark(MJ, SJ, IIIT, SVT, IC03, IC13, IC15, SVTP, and CUTE.) | [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) | LMDB格式,可直接用[lmdb_dataset.py](../../../ppocr/data/lmdb_dataset.py)加载 |
|ICDAR 2015| http://rrc.cvc.uab.es/?ch=4&com=downloads | [train](https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt)/ [test](https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt) |
| 多语言数据集 |[百度网盘](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) 提取码:frgi <br> [google drive](https://drive.google.com/file/d/18cSWX7wXSy4G0tbKJ0d9PuIaiwRLHpjA/view) | 图片下载地址中已包含 |
#### 2.1 ICDAR 2015
ICDAR 2015 数据集可以在上表中链接下载,用于快速验证。也可以从上表中下载 en benchmark 所需的lmdb格式数据集。
下载完图片后从上表中下载转换好的标注文件。
PaddleOCR 也提供了数据格式转换脚本,可以将ICDAR官网 label 转换为PaddleOCR支持的数据格式。 数据转换工具在 `ppocr/utils/gen_label.py`, 这里以训练集为例:
```
# 将官网下载的标签文件转换为 rec_gt_label.txt
python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
```
数据样式格式如下,(a)为原始图片,(b)为每张图片对应的 Ground Truth 文本文件:
![](../../datasets/icdar_rec.png)
## 3. 数据存放路径
PaddleOCR训练数据的默认存储路径是 `PaddleOCR/train_data`,如果您的磁盘上已有数据集,只需创建软链接至数据集目录:
```
# linux and mac os
ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
# windows
mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
```
# 表格识别数据集
- [数据集汇总](#数据集汇总)
- [1. PubTabNet数据集](#1-pubtabnet数据集)
- [2. 好未来表格识别竞赛数据集](#2-好未来表格识别竞赛数据集)
这里整理了常用表格识别数据集,持续更新中,欢迎各位小伙伴贡献数据集~
## 数据集汇总
| 数据集名称 |图片下载地址| PPOCR标注下载地址 |
|---|---|---|
| PubTabNet |https://github.com/ibm-aur-nlp/PubTabNet| jsonl格式,可直接用[pubtab_dataset.py](../../../ppocr/data/pubtab_dataset.py)加载 |
| 好未来表格识别竞赛数据集 |https://ai.100tal.com/dataset| jsonl格式,可直接用[pubtab_dataset.py](../../../ppocr/data/pubtab_dataset.py)加载 |
## 1. PubTabNet数据集
- **数据简介**:PubTabNet数据集的训练集合中包含50万张图像,验证集合中包含0.9万张图像。部分图像可视化如下所示。
<div align="center">
<img src="../../datasets/table_PubTabNet_demo/PMC524509_007_00.png" width="500">
<img src="../../datasets/table_PubTabNet_demo/PMC535543_007_01.png" width="500">
</div>
- **说明**:使用该数据集时,需要遵守[CDLA-Permissive](https://cdla.io/permissive-1-0/)协议。
## 2. 好未来表格识别竞赛数据集
- **数据简介**:好未来表格识别竞赛数据集的训练集合中包含1.6万张图像。验证集未给出可训练的标注。
<div align="center">
<img src="../../datasets/table_tal_demo/1.jpg" width="500">
<img src="../../datasets/table_tal_demo/2.jpg" width="500">
</div>
...@@ -22,7 +22,7 @@ ...@@ -22,7 +22,7 @@
* CCPD-Challenge: 至今在车牌检测识别任务中最有挑战性的一些图片 * CCPD-Challenge: 至今在车牌检测识别任务中最有挑战性的一些图片
* CCPD-NP: 没有安装车牌的新车图片。 * CCPD-NP: 没有安装车牌的新车图片。
![](../datasets/ccpd_demo.png) ![](../../datasets/ccpd_demo.png)
- **下载地址** - **下载地址**
...@@ -46,7 +46,7 @@ ...@@ -46,7 +46,7 @@
* 有效期结束:07/41 * 有效期结束:07/41
* 卡用户拼音:MICHAEL * 卡用户拼音:MICHAEL
![](../datasets/cmb_demo.jpg) ![](../../datasets/cmb_demo.jpg)
- **下载地址**: [https://cdn.kesci.com/cmb2017-2.zip](https://cdn.kesci.com/cmb2017-2.zip) - **下载地址**: [https://cdn.kesci.com/cmb2017-2.zip](https://cdn.kesci.com/cmb2017-2.zip)
...@@ -59,7 +59,7 @@ ...@@ -59,7 +59,7 @@
- **数据简介**: 这是一个数据合成的工具包,可以根据输入的文本,输出验证码图片,使用该工具包生成几张demo图片如下。 - **数据简介**: 这是一个数据合成的工具包,可以根据输入的文本,输出验证码图片,使用该工具包生成几张demo图片如下。
![](../datasets/captcha_demo.png) ![](../../datasets/captcha_demo.png)
- **下载地址**: 该数据集是生成得到,无下载地址。 - **下载地址**: 该数据集是生成得到,无下载地址。
......
...@@ -2,73 +2,30 @@ ...@@ -2,73 +2,30 @@
本节以icdar2015数据集为例,介绍PaddleOCR中检测模型训练、评估、测试的使用方式。 本节以icdar2015数据集为例,介绍PaddleOCR中检测模型训练、评估、测试的使用方式。
- [1. 准备数据和模型](#1--------) - [1. 准备数据和模型](#1-准备数据和模型)
* [1.1 数据准备](#11-----) - [1.1 准备数据集](#11-准备数据集)
* [1.2 下载预训练模型](#12--------) - [1.2 下载预训练模型](#12-下载预训练模型)
- [2. 开始训练](#2-----) - [2. 开始训练](#2-开始训练)
* [2.1 启动训练](#21-----) - [2.1 启动训练](#21-启动训练)
* [2.2 断点训练](#22-----) - [2.2 断点训练](#22-断点训练)
* [2.3 更换Backbone 训练](#23---backbone---) - [2.3 更换Backbone 训练](#23-更换backbone-训练)
* [2.4 混合精度训练](#24---amp---) - [2.4 混合精度训练](#24-混合精度训练)
* [2.5 分布式训练](#25---fleet---) - [2.5 分布式训练](#25-分布式训练)
* [2.6 知识蒸馏训练](#26---distill---) - [2.6 知识蒸馏训练](#26-知识蒸馏训练)
* [2.7 其他训练环境(Windows/macOS/Linux DCU)](#27---other---) - [2.7 其他训练环境](#27-其他训练环境)
- [3. 模型评估与预测](#3--------) - [3. 模型评估与预测](#3-模型评估与预测)
* [3.1 指标评估](#31-----) - [3.1 指标评估](#31-指标评估)
* [3.2 测试检测效果](#32-------) - [3.2 测试检测效果](#32-测试检测效果)
- [4. 模型导出与预测](#4--------) - [4. 模型导出与预测](#4-模型导出与预测)
- [5. FAQ](#5-faq) - [5. FAQ](#5-faq)
<a name="1--------"></a> <a name="1--------"></a>
# 1. 准备数据和模型 # 1. 准备数据和模型
<a name="11-----"></a> ## 1.1 准备数据集
## 1.1 数据准备
icdar2015 TextLocalization数据集是文本检测的数据集,包含1000张训练图像和500张测试图像。 准备数据集可参考 [ocr_datasets](./dataset/ocr_datasets.md)
icdar2015数据集可以从[官网](https://rrc.cvc.uab.es/?ch=4&com=downloads)下载到,首次下载需注册。
注册完成登陆后,下载下图中红色框标出的部分,其中, `Training Set Images`下载的内容保存为`icdar_c4_train_imgs`文件夹下,`Test Set Images` 下载的内容保存为`ch4_test_images`文件夹下
<p align="center">
<img src="../datasets/ic15_location_download.png" align="middle" width = "700"/>
<p align="center">
将下载到的数据集解压到工作目录下,假设解压在 PaddleOCR/train_data/下。另外,PaddleOCR将零散的标注文件整理成单独的标注文件
,您可以通过wget的方式进行下载。
```shell
# 在PaddleOCR路径下
cd PaddleOCR/
wget -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/train_icdar2015_label.txt
wget -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/test_icdar2015_label.txt
```
PaddleOCR 也提供了数据格式转换脚本,可以将官网 label 转换支持的数据格式。 数据转换工具在 `ppocr/utils/gen_label.py`, 这里以训练集为例:
```
# 将官网下载的标签文件转换为 train_icdar2015_label.txt
python gen_label.py --mode="det" --root_path="/path/to/icdar_c4_train_imgs/" \
--input_path="/path/to/ch4_training_localization_transcription_gt" \
--output_label="/path/to/train_icdar2015_label.txt"
```
解压数据集和下载标注文件后,PaddleOCR/train_data/ 有两个文件夹和两个文件,按照如下方式组织icdar2015数据集:
```
/PaddleOCR/train_data/icdar2015/text_localization/
└─ icdar_c4_train_imgs/ icdar数据集的训练数据
└─ ch4_test_images/ icdar数据集的测试数据
└─ train_icdar2015_label.txt icdar数据集的训练标注
└─ test_icdar2015_label.txt icdar数据集的测试标注
```
提供的标注文件格式如下,中间用"\t"分隔:
```
" 图像文件名 json.dumps编码的图像标注信息"
ch4_test_images/img_61.jpg [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]
```
json.dumps编码前的图像标注信息是包含多个字典的list,字典中的 `points` 表示文本框的四个点的坐标(x, y),从左上角的点开始顺时针排列。
`transcription` 表示当前文本框的文字,**当其内容为“###”时,表示该文本框无效,在训练时会跳过。**
如果您想在其他数据集上训练,可以按照上述形式构建标注文件。
<a name="12--------"></a> <a name="12--------"></a>
## 1.2 下载预训练模型 ## 1.2 下载预训练模型
...@@ -220,7 +177,6 @@ Windows平台只支持`单卡`的训练与预测,指定GPU进行训练`set CUD ...@@ -220,7 +177,6 @@ Windows平台只支持`单卡`的训练与预测,指定GPU进行训练`set CUD
- Linux DCU - Linux DCU
DCU设备上运行需要设置环境变量 `export HIP_VISIBLE_DEVICES=0,1,2,3`,其余训练评估预测命令与Linux GPU完全相同。 DCU设备上运行需要设置环境变量 `export HIP_VISIBLE_DEVICES=0,1,2,3`,其余训练评估预测命令与Linux GPU完全相同。
<a name="3--------"></a> <a name="3--------"></a>
# 3. 模型评估与预测 # 3. 模型评估与预测
......
...@@ -27,6 +27,7 @@ ...@@ -27,6 +27,7 @@
<a name="1-数据准备"></a> <a name="1-数据准备"></a>
# 1. 数据准备 # 1. 数据准备
### 1.1 准备数据集
PaddleOCR 支持两种数据格式: PaddleOCR 支持两种数据格式:
- `lmdb` 用于训练以lmdb格式存储的数据集(LMDBDataSet); - `lmdb` 用于训练以lmdb格式存储的数据集(LMDBDataSet);
......
...@@ -81,13 +81,13 @@ Optimizer: ...@@ -81,13 +81,13 @@ Optimizer:
- 检测: - 检测:
- 英文数据集,ICDAR2015 - 英文数据集,ICDAR2015
- 中文数据集,LSVT街景数据集训练数据3w张图片 - 中文数据集,LSVT街景数据集训练数据3w张图片
- 识别: - 识别:
- 英文数据集,MJSynth和SynthText合成数据,数据量上千万。 - 英文数据集,MJSynth和SynthText合成数据,数据量上千万。
- 中文数据集,LSVT街景数据集根据真值将图crop出来,并进行位置校准,总共30w张图像。此外基于LSVT的语料,合成数据500w。 - 中文数据集,LSVT街景数据集根据真值将图crop出来,并进行位置校准,总共30w张图像。此外基于LSVT的语料,合成数据500w。
- 小语种数据集,使用不同语料和字体,分别生成了100w合成数据集,并使用ICDAR-MLT作为验证集。 - 小语种数据集,使用不同语料和字体,分别生成了100w合成数据集,并使用ICDAR-MLT作为验证集。
其中,公开数据集都是开源的,用户可自行搜索下载,也可参考[中文数据集](./datasets.md),合成数据暂不开源,用户可使用开源合成工具自行合成,可参考的合成工具包括[text_renderer](https://github.com/Sanster/text_renderer)[SynthText](https://github.com/ankush-me/SynthText)[TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) 等。 其中,公开数据集都是开源的,用户可自行搜索下载,也可参考[中文数据集](dataset/datasets.md),合成数据暂不开源,用户可使用开源合成工具自行合成,可参考的合成工具包括[text_renderer](https://github.com/Sanster/text_renderer)[SynthText](https://github.com/ankush-me/SynthText)[TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) 等。
<a name="垂类场景"></a> <a name="垂类场景"></a>
### 3.2 垂类场景 ### 3.2 垂类场景
...@@ -120,17 +120,17 @@ PaddleOCR主要聚焦通用OCR,如果有垂类需求,您可以用PaddleOCR+ ...@@ -120,17 +120,17 @@ PaddleOCR主要聚焦通用OCR,如果有垂类需求,您可以用PaddleOCR+
**Q**:训练CRNN识别时,如何选择合适的网络输入shape? **Q**:训练CRNN识别时,如何选择合适的网络输入shape?
A:一般高度采用32,最长宽度的选择,有两种方法: A:一般高度采用32,最长宽度的选择,有两种方法:
(1)统计训练样本图像的宽高比分布。最大宽高比的选取考虑满足80%的训练样本。 (1)统计训练样本图像的宽高比分布。最大宽高比的选取考虑满足80%的训练样本。
(2)统计训练样本文字数目。最长字符数目的选取考虑满足80%的训练样本。然后中文字符长宽比近似认为是1,英文认为3:1,预估一个最长宽度。 (2)统计训练样本文字数目。最长字符数目的选取考虑满足80%的训练样本。然后中文字符长宽比近似认为是1,英文认为3:1,预估一个最长宽度。
**Q**:识别训练时,训练集精度已经到达90了,但验证集精度一直在70,涨不上去怎么办? **Q**:识别训练时,训练集精度已经到达90了,但验证集精度一直在70,涨不上去怎么办?
A:训练集精度90,测试集70多的话,应该是过拟合了,有两个可尝试的方法: A:训练集精度90,测试集70多的话,应该是过拟合了,有两个可尝试的方法:
(1)加入更多的增广方式或者调大增广prob的[概率](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/rec_img_aug.py#L341),默认为0.4。 (1)加入更多的增广方式或者调大增广prob的[概率](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/rec_img_aug.py#L341),默认为0.4。
(2)调大系统的[l2 dcay值](https://github.com/PaddlePaddle/PaddleOCR/blob/a501603d54ff5513fc4fc760319472e59da25424/configs/rec/ch_ppocr_v1.1/rec_chinese_lite_train_v1.1.yml#L47) (2)调大系统的[l2 dcay值](https://github.com/PaddlePaddle/PaddleOCR/blob/a501603d54ff5513fc4fc760319472e59da25424/configs/rec/ch_ppocr_v1.1/rec_chinese_lite_train_v1.1.yml#L47)
**Q**: 识别模型训练时,loss能正常下降,但acc一直为0 **Q**: 识别模型训练时,loss能正常下降,但acc一直为0
...@@ -141,12 +141,11 @@ PaddleOCR主要聚焦通用OCR,如果有垂类需求,您可以用PaddleOCR+ ...@@ -141,12 +141,11 @@ PaddleOCR主要聚焦通用OCR,如果有垂类需求,您可以用PaddleOCR+
*** ***
具体的训练教程可点击下方链接跳转: 具体的训练教程可点击下方链接跳转:
- [文本检测模型训练](./detection.md) - [文本检测模型训练](./detection.md)
- [文本识别模型训练](./recognition.md) - [文本识别模型训练](./recognition.md)
- [文本方向分类器训练](./angle_class.md) - [文本方向分类器训练](./angle_class.md)
- [知识蒸馏](./knowledge_distillation.md) - [知识蒸馏](./knowledge_distillation.md)
...@@ -22,7 +22,7 @@ ...@@ -22,7 +22,7 @@
- 2020.7.15 整理OCR相关数据集、常用数据标注以及合成工具 - 2020.7.15 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](./config.md) - 2020.7.9 添加数据增强、学习率衰减策略,具体参考[配置文件](./config.md)
- 2020.6.8 添加[数据集](./datasets.md),并保持持续更新 - 2020.6.8 添加[数据集](dataset/datasets.md),并保持持续更新
- 2020.6.5 支持 `attetnion` 模型导出 `inference_model` - 2020.6.5 支持 `attetnion` 模型导出 `inference_model`
- 2020.6.5 支持单独预测识别时,输出结果得分 - 2020.6.5 支持单独预测识别时,输出结果得分
- 2020.5.30 提供超轻量级中文OCR在线体验 - 2020.5.30 提供超轻量级中文OCR在线体验
......
...@@ -42,7 +42,7 @@ At present, the open source model, dataset and magnitude are as follows: ...@@ -42,7 +42,7 @@ At present, the open source model, dataset and magnitude are as follows:
English dataset: MJSynth and SynthText synthetic dataset, the amount of data is tens of millions. English dataset: MJSynth and SynthText synthetic dataset, the amount of data is tens of millions.
Chinese dataset: LSVT street view dataset with cropped text area, a total of 30w images. In addition, the synthesized data based on LSVT corpus is 500w. Chinese dataset: LSVT street view dataset with cropped text area, a total of 30w images. In addition, the synthesized data based on LSVT corpus is 500w.
Among them, the public datasets are opensourced, users can search and download by themselves, or refer to [Chinese data set](./datasets_en.md), synthetic data is not opensourced, users can use open-source synthesis tools to synthesize data themselves. Current available synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator), etc. Among them, the public datasets are opensourced, users can search and download by themselves, or refer to [Chinese data set](dataset/datasets_en.md), synthetic data is not opensourced, users can use open-source synthesis tools to synthesize data themselves. Current available synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator), etc.
10. **Error in using the model with TPS module for prediction** 10. **Error in using the model with TPS module for prediction**
Error message: Input(X) dims[3] and Input(Grid) dims[2] should be equal, but received X dimension[3]\(108) != Grid dimension[2]\(100) Error message: Input(X) dims[3] and Input(Grid) dims[2] should be equal, but received X dimension[3]\(108) != Grid dimension[2]\(100)
......
# FCENet
- [1. Introduction](#1)
- [2. Environment](#2)
- [3. Model Training / Evaluation / Prediction](#3)
- [3.1 Training](#3-1)
- [3.2 Evaluation](#3-2)
- [3.3 Prediction](#3-3)
- [4. Inference and Deployment](#4)
- [4.1 Python Inference](#4-1)
- [4.2 C++ Inference](#4-2)
- [4.3 Serving](#4-3)
- [4.4 More](#4-4)
- [5. FAQ](#5)
<a name="1"></a>
## 1. Introduction
Paper:
> [Fourier Contour Embedding for Arbitrary-Shaped Text Detection](https://arxiv.org/abs/2104.10442)
> Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang
> CVPR, 2021
On the CTW1500 dataset, the text detection result is as follows:
|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
| --- | --- | --- | --- | --- | --- | --- |
| FCE | ResNet50_dcn | [configs/det/det_r50_vd_dcn_fce_ctw.yml](../../configs/det/det_r50_vd_dcn_fce_ctw.yml)| 88.39%|82.18%|85.27%|[trained model](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar)|
<a name="2"></a>
## 2. Environment
Please prepare your environment referring to [prepare the environment](./environment_en.md) and [clone the repo](./clone_en.md).
<a name="3"></a>
## 3. Model Training / Evaluation / Prediction
The above FCE model is trained using the CTW1500 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](./dataset/ocr_datasets_en.md).
After the data download is complete, please refer to [Text Detection Training Tutorial](./detection.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
<a name="4"></a>
## 4. Inference and Deployment
<a name="4-1"></a>
### 4.1 Python Inference
First, convert the model saved in the FCE text detection training process into an inference model. Taking the model based on the Resnet50_vd_dcn backbone network and trained on the CTW1500 English dataset as example ([model download link](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar)), you can use the following command to convert:
```shell
python3 tools/export_model.py -c configs/det/det_r50_vd_dcn_fce_ctw.yml -o Global.pretrained_model=./det_r50_dcn_fce_ctw_v2.0_train/best_accuracy Global.save_inference_dir=./inference/det_fce
```
FCE text detection model inference, to perform non-curved text detection, you can run the following commands:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=quad
```
The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
![](../imgs_results/det_res_img_10_fce.jpg)
If you want to perform curved text detection, you can execute the following command:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=poly
```
The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
![](../imgs_results/det_res_img623_fce.jpg)
**Note**: Since the CTW1500 dataset has only 1,000 training images, mainly for English scenes, the above model has very poor detection result on Chinese or curved text images.
<a name="4-2"></a>
### 4.2 C++ Inference
Since the post-processing is not written in CPP, the FCE text detection model does not support CPP inference.
<a name="4-3"></a>
### 4.3 Serving
Not supported
<a name="4-4"></a>
### 4.4 More
Not supported
<a name="5"></a>
## 5. FAQ
## Citation
```bibtex
@InProceedings{zhu2021fourier,
title={Fourier Contour Embedding for Arbitrary-Shaped Text Detection},
author={Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang},
year={2021},
booktitle = {CVPR}
}
```
# PSENet
- [1. Introduction](#1)
- [2. Environment](#2)
- [3. Model Training / Evaluation / Prediction](#3)
- [3.1 Training](#3-1)
- [3.2 Evaluation](#3-2)
- [3.3 Prediction](#3-3)
- [4. Inference and Deployment](#4)
- [4.1 Python Inference](#4-1)
- [4.2 C++ Inference](#4-2)
- [4.3 Serving](#4-3)
- [4.4 More](#4-4)
- [5. FAQ](#5)
<a name="1"></a>
## 1. Introduction
Paper:
> [Shape robust text detection with progressive scale expansion network](https://arxiv.org/abs/1903.12473)
> Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai
> CVPR, 2019
On the ICDAR2015 dataset, the text detection result is as follows:
|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
| --- | --- | --- | --- | --- | --- | --- |
|PSE| ResNet50_vd | [configs/det/det_r50_vd_pse.yml](../../configs/det/det_r50_vd_pse.yml)| 85.81% |79.53%|82.55%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
|PSE| MobileNetV3| [configs/det/det_mv3_pse.yml](../../configs/det/det_mv3_pse.yml) | 82.20% |70.48%|75.89%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
<a name="2"></a>
## 2. Environment
Please prepare your environment referring to [prepare the environment](./environment_en.md) and [clone the repo](./clone_en.md).
<a name="3"></a>
## 3. Model Training / Evaluation / Prediction
The above PSE model is trained using the ICDAR2015 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](./dataset/ocr_datasets_en.md).
After the data download is complete, please refer to [Text Detection Training Tutorial](./detection.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
<a name="4"></a>
## 4. Inference and Deployment
<a name="4-1"></a>
### 4.1 Python Inference
First, convert the model saved in the PSE text detection training process into an inference model. Taking the model based on the Resnet50_vd backbone network and trained on the ICDAR2015 English dataset as example ([model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)), you can use the following command to convert:
```shell
python3 tools/export_model.py -c configs/det/det_r50_vd_pse.yml -o Global.pretrained_model=./det_r50_vd_pse_v2.0_train/best_accuracy Global.save_inference_dir=./inference/det_pse
```
PSE text detection model inference, to perform non-curved text detection, you can run the following commands:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=quad
```
The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
![](../imgs_results/det_res_img_10_pse.jpg)
If you want to perform curved text detection, you can execute the following command:
```shell
python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=poly
```
The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
![](../imgs_results/det_res_img_10_pse_poly.jpg)
**Note**: Since the ICDAR2015 dataset has only 1,000 training images, mainly for English scenes, the above model has very poor detection result on Chinese or curved text images.
<a name="4-2"></a>
### 4.2 C++ Inference
Since the post-processing is not written in CPP, the PSE text detection model does not support CPP inference.
<a name="4-3"></a>
### 4.3 Serving
Not supported
<a name="4-4"></a>
### 4.4 More
Not supported
<a name="5"></a>
## 5. FAQ
## Citation
```bibtex
@inproceedings{wang2019shape,
title={Shape robust text detection with progressive scale expansion network},
author={Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={9336--9345},
year={2019}
}
```
...@@ -12,12 +12,12 @@ In addition to opensource data, users can also use synthesis tools to synthesize ...@@ -12,12 +12,12 @@ In addition to opensource data, users can also use synthesis tools to synthesize
#### 1. ICDAR2019-LSVT #### 1. ICDAR2019-LSVT
- **Data sources**:https://ai.baidu.com/broad/introduction?dataset=lsvt - **Data sources**:https://ai.baidu.com/broad/introduction?dataset=lsvt
- **Introduction**: A total of 45w Chinese street view images, including 5w (2w test + 3w training) fully labeled data (text coordinates + text content), 40w weakly labeled data (text content only), as shown in the following figure: - **Introduction**: A total of 45w Chinese street view images, including 5w (2w test + 3w training) fully labeled data (text coordinates + text content), 40w weakly labeled data (text content only), as shown in the following figure:
![](../datasets/LSVT_1.jpg) ![](../../datasets/LSVT_1.jpg)
(a) Fully labeled data (a) Fully labeled data
![](../datasets/LSVT_2.jpg) ![](../../datasets/LSVT_2.jpg)
(b) Weakly labeled data (b) Weakly labeled data
- **Download link**:https://ai.baidu.com/broad/download?dataset=lsvt - **Download link**:https://ai.baidu.com/broad/download?dataset=lsvt
...@@ -25,7 +25,7 @@ In addition to opensource data, users can also use synthesis tools to synthesize ...@@ -25,7 +25,7 @@ In addition to opensource data, users can also use synthesis tools to synthesize
#### 2. ICDAR2017-RCTW-17 #### 2. ICDAR2017-RCTW-17
- **Data sources**:https://rctw.vlrlab.net/ - **Data sources**:https://rctw.vlrlab.net/
- **Introduction**:It contains 12000 + images, most of them are collected in the wild through mobile camera. Some are screenshots. These images show a variety of scenes, including street views, posters, menus, indoor scenes and screenshots of mobile applications. - **Introduction**:It contains 12000 + images, most of them are collected in the wild through mobile camera. Some are screenshots. These images show a variety of scenes, including street views, posters, menus, indoor scenes and screenshots of mobile applications.
![](../datasets/rctw.jpg) ![](../../datasets/rctw.jpg)
- **Download link**:https://rctw.vlrlab.net/dataset/ - **Download link**:https://rctw.vlrlab.net/dataset/
<a name="中文街景文字识别"></a> <a name="中文街景文字识别"></a>
...@@ -33,9 +33,9 @@ In addition to opensource data, users can also use synthesis tools to synthesize ...@@ -33,9 +33,9 @@ In addition to opensource data, users can also use synthesis tools to synthesize
- **Data sources**:https://aistudio.baidu.com/aistudio/competition/detail/8 - **Data sources**:https://aistudio.baidu.com/aistudio/competition/detail/8
- **Introduction**:A total of 290000 pictures are included, of which 210000 are used as training sets (with labels) and 80000 are used as test sets (without labels). The dataset is collected from the Chinese street view, and is formed by by cutting out the text line area (such as shop signs, landmarks, etc.) in the street view picture. All the images are preprocessed: by using affine transform, the text area is proportionally mapped to a picture with a height of 48 pixels, as shown in the figure: - **Introduction**:A total of 290000 pictures are included, of which 210000 are used as training sets (with labels) and 80000 are used as test sets (without labels). The dataset is collected from the Chinese street view, and is formed by by cutting out the text line area (such as shop signs, landmarks, etc.) in the street view picture. All the images are preprocessed: by using affine transform, the text area is proportionally mapped to a picture with a height of 48 pixels, as shown in the figure:
![](../datasets/ch_street_rec_1.png) ![](../../datasets/ch_street_rec_1.png)
(a) Label: 魅派集成吊顶 (a) Label: 魅派集成吊顶
![](../datasets/ch_street_rec_2.png) ![](../../datasets/ch_street_rec_2.png)
(b) Label: 母婴用品连锁 (b) Label: 母婴用品连锁
- **Download link** - **Download link**
https://aistudio.baidu.com/aistudio/datasetdetail/8429 https://aistudio.baidu.com/aistudio/datasetdetail/8429
...@@ -49,13 +49,13 @@ https://aistudio.baidu.com/aistudio/datasetdetail/8429 ...@@ -49,13 +49,13 @@ https://aistudio.baidu.com/aistudio/datasetdetail/8429
- 5990 characters including Chinese characters, English letters, numbers and punctuation(Characters set: https://github.com/YCG09/chinese_ocr/blob/master/train/char_std_5990.txt ) - 5990 characters including Chinese characters, English letters, numbers and punctuation(Characters set: https://github.com/YCG09/chinese_ocr/blob/master/train/char_std_5990.txt )
- Each sample is fixed with 10 characters, and the characters are randomly intercepted from the sentences in the corpus - Each sample is fixed with 10 characters, and the characters are randomly intercepted from the sentences in the corpus
- Image resolution is 280x32 - Image resolution is 280x32
![](../datasets/ch_doc1.jpg) ![](../../datasets/ch_doc1.jpg)
![](../datasets/ch_doc3.jpg) ![](../../datasets/ch_doc3.jpg)
- **Download link**:https://pan.baidu.com/s/1QkI7kjah8SPHwOQ40rS1Pw (Password: lu7m) - **Download link**:https://pan.baidu.com/s/1QkI7kjah8SPHwOQ40rS1Pw (Password: lu7m)
<a name="ICDAR2019-ArT"></a> <a name="ICDAR2019-ArT"></a>
#### 5、ICDAR2019-ArT #### 5、ICDAR2019-ArT
- **Data source**:https://ai.baidu.com/broad/introduction?dataset=art - **Data source**:https://ai.baidu.com/broad/introduction?dataset=art
- **Introduction**:It includes 10166 images, 5603 in training sets and 4563 in test sets. It is composed of three parts: total text, scut-ctw1500 and Baidu curved scene text, including text with various shapes such as horizontal, multi-directional and curved. - **Introduction**:It includes 10166 images, 5603 in training sets and 4563 in test sets. It is composed of three parts: total text, scut-ctw1500 and Baidu curved scene text, including text with various shapes such as horizontal, multi-directional and curved.
![](../datasets/ArT.jpg) ![](../../datasets/ArT.jpg)
- **Download link**:https://ai.baidu.com/broad/download?dataset=art - **Download link**:https://ai.baidu.com/broad/download?dataset=art
...@@ -9,7 +9,7 @@ Here we have sorted out the commonly used handwritten OCR dataset datasets, whic ...@@ -9,7 +9,7 @@ Here we have sorted out the commonly used handwritten OCR dataset datasets, whic
- **Data introduction**: - **Data introduction**:
* It includes online and offline handwritten data,`HWDB1.0~1.2` has totally 3895135 handwritten single character samples, which belong to 7356 categories (7185 Chinese characters and 171 English letters, numbers and symbols);`HWDB2.0~2.2` has totally 5091 pages of images, which are divided into 52230 text lines and 1349414 words. All text and text samples are stored as grayscale images. Some sample words are shown below. * It includes online and offline handwritten data,`HWDB1.0~1.2` has totally 3895135 handwritten single character samples, which belong to 7356 categories (7185 Chinese characters and 171 English letters, numbers and symbols);`HWDB2.0~2.2` has totally 5091 pages of images, which are divided into 52230 text lines and 1349414 words. All text and text samples are stored as grayscale images. Some sample words are shown below.
![](../datasets/CASIA_0.jpg) ![](../../datasets/CASIA_0.jpg)
- **Download address**:http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html - **Download address**:http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html
- **使用建议**:Data for single character, white background, can form a large number of text lines for training. White background can be processed into transparent state, which is convenient to add various backgrounds. For the case of semantic needs, it is suggested to extract single character from real corpus to form text lines. - **使用建议**:Data for single character, white background, can form a large number of text lines for training. White background can be processed into transparent state, which is convenient to add various backgrounds. For the case of semantic needs, it is suggested to extract single character from real corpus to form text lines.
...@@ -22,7 +22,7 @@ Here we have sorted out the commonly used handwritten OCR dataset datasets, whic ...@@ -22,7 +22,7 @@ Here we have sorted out the commonly used handwritten OCR dataset datasets, whic
- **Data introduction**: NIST19 dataset is suitable for handwritten document and character recognition model training. It is extracted from the handwritten sample form of 3600 authors and contains 810000 character images in total. Nine of them are shown below. - **Data introduction**: NIST19 dataset is suitable for handwritten document and character recognition model training. It is extracted from the handwritten sample form of 3600 authors and contains 810000 character images in total. Nine of them are shown below.
![](../datasets/nist_demo.png) ![](../../datasets/nist_demo.png)
- **Download address**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19) - **Download address**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19)
# OCR datasets
- [1. Text detection](#1-text-detection)
- [1.1 PaddleOCR text detection format annotation](#11-paddleocr-text-detection-format-annotation)
- [1.2 Public dataset](#12-public-dataset)
- [1.2.1 ICDAR 2015](#121-icdar-2015)
- [2. Text recognition](#2-text-recognition)
- [2.1 PaddleOCR text recognition format annotation](#21-paddleocr-text-recognition-format-annotation)
- [2.2 Public dataset](#22-public-dataset)
- [2.1 ICDAR 2015](#21-icdar-2015)
- [3. Data storage path](#3-data-storage-path)
Here is a list of public datasets commonly used in OCR, which are being continuously updated. Welcome to contribute datasets~
## 1. Text detection
### 1.1 PaddleOCR text detection format annotation
The annotation file formats supported by the PaddleOCR text detection algorithm are as follows, separated by "\t":
```
" Image file name Image annotation information encoded by json.dumps"
ch4_test_images/img_61.jpg [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]
```
The image annotation after **json.dumps()** encoding is a list containing multiple dictionaries.
The `points` in the dictionary represent the coordinates (x, y) of the four points of the text box, arranged clockwise from the point at the upper left corner.
`transcription` represents the text of the current text box. **When its content is "###" it means that the text box is invalid and will be skipped during training.**
If you want to train PaddleOCR on other datasets, please build the annotation file according to the above format.
### 1.2 Public dataset
| dataset | Image download link | PaddleOCR format annotation download link |
|---|---|---|
| ICDAR 2015 | https://rrc.cvc.uab.es/?ch=4&com=downloads | [train](https://paddleocr.bj.bcebos.com/dataset/train_icdar2015_label.txt) / [test](https://paddleocr.bj.bcebos.com/dataset/test_icdar2015_label.txt) |
| ctw1500 | https://paddleocr.bj.bcebos.com/dataset/ctw1500.zip | Included in the downloaded image zip |
| total text | https://paddleocr.bj.bcebos.com/dataset/total_text.tar | Included in the downloaded image zip |
#### 1.2.1 ICDAR 2015
The icdar2015 dataset contains train set which has 1000 images obtained with wearable cameras and test set which has 500 images obtained with wearable cameras. The icdar2015 dataset can be downloaded from the link in the table above. Registration is required for downloading.
After registering and logging in, download the part marked in the red box in the figure below. And, the content downloaded by `Training Set Images` should be saved as the folder `icdar_c4_train_imgs`, and the content downloaded by `Test Set Images` is saved as the folder `ch4_test_images`
<p align="center">
<img src="../../datasets/ic15_location_download.png" align="middle" width = "700"/>
<p align="center">
Decompress the downloaded dataset to the working directory, assuming it is decompressed under PaddleOCR/train_data/. Then download the PaddleOCR format annotation file from the table above.
PaddleOCR also provides a data format conversion script, which can convert the official website label to the PaddleOCR format. The data conversion tool is in `ppocr/utils/gen_label.py`, here is the training set as an example:
```
# Convert the label file downloaded from the official website to train_icdar2015_label.txt
python gen_label.py --mode="det" --root_path="/path/to/icdar_c4_train_imgs/" \
--input_path="/path/to/ch4_training_localization_transcription_gt" \
--output_label="/path/to/train_icdar2015_label.txt"
```
After decompressing the data set and downloading the annotation file, PaddleOCR/train_data/ has two folders and two files, which are:
```
/PaddleOCR/train_data/icdar2015/text_localization/
└─ icdar_c4_train_imgs/ Training data of icdar dataset
└─ ch4_test_images/ Testing data of icdar dataset
└─ train_icdar2015_label.txt Training annotation of icdar dataset
└─ test_icdar2015_label.txt Test annotation of icdar dataset
```
## 2. Text recognition
### 2.1 PaddleOCR text recognition format annotation
The text recognition algorithm in PaddleOCR supports two data formats:
- `lmdb` is used to train data sets stored in lmdb format, use [lmdb_dataset.py](../../../ppocr/data/lmdb_dataset.py) to load;
- `common dataset` is used to train data sets stored in text files, use [simple_dataset.py](../../../ppocr/data/simple_dataset.py) to load.
If you want to use your own data for training, please refer to the following to organize your data.
- Training set
It is recommended to put the training images in the same folder, and use a txt file (rec_gt_train.txt) to store the image path and label. The contents of the txt file are as follows:
* Note: by default, the image path and image label are split with \t, if you use other methods to split, it will cause training error
```
" Image file name Image annotation "
train_data/rec/train/word_001.jpg 简单可依赖
train_data/rec/train/word_002.jpg 用科技让复杂的世界更简单
...
```
The final training set should have the following file structure:
```
|-train_data
|-rec
|- rec_gt_train.txt
|- train
|- word_001.png
|- word_002.jpg
|- word_003.jpg
| ...
```
- Test set
Similar to the training set, the test set also needs to be provided a folder containing all images (test) and a rec_gt_test.txt. The structure of the test set is as follows:
```
|-train_data
|-rec
|-ic15_data
|- rec_gt_test.txt
|- test
|- word_001.jpg
|- word_002.jpg
|- word_003.jpg
| ...
```
### 2.2 Public dataset
| dataset | Image download link | PaddleOCR format annotation download link |
|---|---|---|
| en benchmark(MJ, SJ, IIIT, SVT, IC03, IC13, IC15, SVTP, and CUTE.) | [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) | LMDB format, which can be loaded directly with [lmdb_dataset.py](../../../ppocr/data/lmdb_dataset.py) |
|ICDAR 2015| http://rrc.cvc.uab.es/?ch=4&com=downloads | [train](https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt)/ [test](https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt) |
| Multilingual datasets |[Baidu network disk](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) Extraction code: frgi <br> [google drive](https://drive.google.com/file/d/18cSWX7wXSy4G0tbKJ0d9PuIaiwRLHpjA/view) | Included in the downloaded image zip |
#### 2.1 ICDAR 2015
The ICDAR 2015 dataset can be downloaded from the link in the table above for quick validation. The lmdb format dataset required by en benchmark can also be downloaded from the table above.
Then download the PaddleOCR format annotation file from the table above.
PaddleOCR also provides a data format conversion script, which can convert the ICDAR official website label to the data format supported by PaddleOCR. The data conversion tool is in `ppocr/utils/gen_label.py`, here is the training set as an example:
```
# Convert the label file downloaded from the official website to rec_gt_label.txt
python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
```
The data format is as follows, (a) is the original picture, (b) is the Ground Truth text file corresponding to each picture:
![](../../datasets/icdar_rec.png)
## 3. Data storage path
The default storage path for PaddleOCR training data is `PaddleOCR/train_data`, if you already have a dataset on your disk, just create a soft link to the dataset directory:
```
# linux and mac os
ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
# windows
mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
```
# Table Recognition Datasets
- [Dataset Summary](#dataset-summary)
- [1. PubTabNet](#1-pubtabnet)
- [2. TAL Table Recognition Competition Dataset](#2-tal-table-recognition-competition-dataset)
Here are the commonly used table recognition datasets, which are being updated continuously. Welcome to contribute datasets~
## Dataset Summary
| dataset | Image download link | PPOCR format annotation download link |
|---|---|---|
| PubTabNet |https://github.com/ibm-aur-nlp/PubTabNet| jsonl format, which can be loaded directly with [pubtab_dataset.py](../../../ppocr/data/pubtab_dataset.py) |
| TAL Table Recognition Competition Dataset |https://ai.100tal.com/dataset| jsonl format, which can be loaded directly with [pubtab_dataset.py](../../../ppocr/data/pubtab_dataset.py) |
## 1. PubTabNet
- **Data Introduction**:The training set of the PubTabNet dataset contains 500,000 images and the validation set contains 9000 images. Part of the image visualization is shown below.
<div align="center">
<img src="../../datasets/table_PubTabNet_demo/PMC524509_007_00.png" width="500">
<img src="../../datasets/table_PubTabNet_demo/PMC535543_007_01.png" width="500">
</div>
- **illustrate**:When using this dataset, the [CDLA-Permissive](https://cdla.io/permissive-1-0/) protocol is required.
## 2. TAL Table Recognition Competition Dataset
- **Data Introduction**:The training set of the TAL table recognition competition dataset contains 16,000 images. The validation set does not give trainable annotations.
<div align="center">
<img src="../../datasets/table_tal_demo/1.jpg" width="500">
<img src="../../datasets/table_tal_demo/2.jpg" width="500">
</div>
...@@ -22,7 +22,7 @@ Here we have sorted out the commonly used vertical multi-language OCR dataset da ...@@ -22,7 +22,7 @@ Here we have sorted out the commonly used vertical multi-language OCR dataset da
* CCPD-Challenge: So far, some of the most challenging images in license plate detection and recognition tasks * CCPD-Challenge: So far, some of the most challenging images in license plate detection and recognition tasks
* CCPD-NP: Pictures of new cars without license plates. * CCPD-NP: Pictures of new cars without license plates.
![](../datasets/ccpd_demo.png) ![](../../datasets/ccpd_demo.png)
- **Download address** - **Download address**
...@@ -46,7 +46,7 @@ Here we have sorted out the commonly used vertical multi-language OCR dataset da ...@@ -46,7 +46,7 @@ Here we have sorted out the commonly used vertical multi-language OCR dataset da
* End of validity: 07/41 * End of validity: 07/41
* Chinese phonetic alphabet of card users: MICHAEL * Chinese phonetic alphabet of card users: MICHAEL
![](../datasets/cmb_demo.jpg) ![](../../datasets/cmb_demo.jpg)
- **Download address**: [https://cdn.kesci.com/cmb2017-2.zip](https://cdn.kesci.com/cmb2017-2.zip) - **Download address**: [https://cdn.kesci.com/cmb2017-2.zip](https://cdn.kesci.com/cmb2017-2.zip)
...@@ -59,7 +59,7 @@ Here we have sorted out the commonly used vertical multi-language OCR dataset da ...@@ -59,7 +59,7 @@ Here we have sorted out the commonly used vertical multi-language OCR dataset da
- **Data introduction**: This is a toolkit for data synthesis. You can output captcha images according to the input text. Use the toolkit to generate several demo images as follows. - **Data introduction**: This is a toolkit for data synthesis. You can output captcha images according to the input text. Use the toolkit to generate several demo images as follows.
![](../datasets/captcha_demo.png) ![](../../datasets/captcha_demo.png)
- **Download address**: The dataset is generated and has no download address. - **Download address**: The dataset is generated and has no download address.
......
...@@ -2,9 +2,9 @@ ...@@ -2,9 +2,9 @@
This section uses the icdar2015 dataset as an example to introduce the training, evaluation, and testing of the detection model in PaddleOCR. This section uses the icdar2015 dataset as an example to introduce the training, evaluation, and testing of the detection model in PaddleOCR.
- [1. Data and Weights Preparation](#1-data-and-weights-preparatio) - [1. Data and Weights Preparation](#1-data-and-weights-preparation)
* [1.1 Data Preparation](#11-data-preparation) - [1.1 Data Preparation](#11-data-preparation)
* [1.2 Download Pre-trained Model](#12-download-pretrained-model) - [1.2 Download Pre-trained Model](#12-download-pre-trained-model)
- [2. Training](#2-training) - [2. Training](#2-training)
* [2.1 Start Training](#21-start-training) * [2.1 Start Training](#21-start-training)
* [2.2 Load Trained Model and Continue Training](#22-load-trained-model-and-continue-training) * [2.2 Load Trained Model and Continue Training](#22-load-trained-model-and-continue-training)
...@@ -14,8 +14,8 @@ This section uses the icdar2015 dataset as an example to introduce the training, ...@@ -14,8 +14,8 @@ This section uses the icdar2015 dataset as an example to introduce the training,
* [2.6 Training with knowledge distillation](#26) * [2.6 Training with knowledge distillation](#26)
* [2.7 Training on other platform(Windows/macOS/Linux DCU)](#27) * [2.7 Training on other platform(Windows/macOS/Linux DCU)](#27)
- [3. Evaluation and Test](#3-evaluation-and-test) - [3. Evaluation and Test](#3-evaluation-and-test)
* [3.1 Evaluation](#31-evaluation) - [3.1 Evaluation](#31-evaluation)
* [3.2 Test](#32-test) - [3.2 Test](#32-test)
- [4. Inference](#4-inference) - [4. Inference](#4-inference)
- [5. FAQ](#5-faq) - [5. FAQ](#5-faq)
...@@ -23,45 +23,7 @@ This section uses the icdar2015 dataset as an example to introduce the training, ...@@ -23,45 +23,7 @@ This section uses the icdar2015 dataset as an example to introduce the training,
### 1.1 Data Preparation ### 1.1 Data Preparation
The icdar2015 dataset contains train set which has 1000 images obtained with wearable cameras and test set which has 500 images obtained with wearable cameras. The icdar2015 can be obtained from [official website](https://rrc.cvc.uab.es/?ch=4&com=downloads). Registration is required for downloading. To prepare datasets, refer to [ocr_datasets](./dataset/ocr_datasets_en.md) .
After registering and logging in, download the part marked in the red box in the figure below. And, the content downloaded by `Training Set Images` should be saved as the folder `icdar_c4_train_imgs`, and the content downloaded by `Test Set Images` is saved as the folder `ch4_test_images`
<p align="center">
<img src="../datasets/ic15_location_download.png" align="middle" width = "700"/>
<p align="center">
Decompress the downloaded dataset to the working directory, assuming it is decompressed under PaddleOCR/train_data/. In addition, PaddleOCR organizes many scattered annotation files into two separate annotation files for train and test respectively, which can be downloaded by wget:
```shell
# Under the PaddleOCR path
cd PaddleOCR/
wget -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/train_icdar2015_label.txt
wget -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/test_icdar2015_label.txt
```
After decompressing the data set and downloading the annotation file, PaddleOCR/train_data/ has two folders and two files, which are:
```
/PaddleOCR/train_data/icdar2015/text_localization/
└─ icdar_c4_train_imgs/ Training data of icdar dataset
└─ ch4_test_images/ Testing data of icdar dataset
└─ train_icdar2015_label.txt Training annotation of icdar dataset
└─ test_icdar2015_label.txt Test annotation of icdar dataset
```
The provided annotation file format is as follow, separated by "\t":
```
" Image file name Image annotation information encoded by json.dumps"
ch4_test_images/img_61.jpg [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]
```
The image annotation after **json.dumps()** encoding is a list containing multiple dictionaries.
The `points` in the dictionary represent the coordinates (x, y) of the four points of the text box, arranged clockwise from the point at the upper left corner.
`transcription` represents the text of the current text box. **When its content is "###" it means that the text box is invalid and will be skipped during training.**
If you want to train PaddleOCR on other datasets, please build the annotation file according to the above format.
### 1.2 Download Pre-trained Model ### 1.2 Download Pre-trained Model
......
...@@ -24,111 +24,14 @@ ...@@ -24,111 +24,14 @@
<a name="DATA_PREPARATION"></a> <a name="DATA_PREPARATION"></a>
## 1. Data Preparation ## 1. Data Preparation
### 1.1 DataSet Preparation
PaddleOCR supports two data formats: To prepare datasets, refer to [ocr_datasets](./dataset/ocr_datasets.md) .
- `LMDB` is used to train data sets stored in lmdb format(LMDBDataSet);
- `general data` is used to train data sets stored in text files(SimpleDataSet):
Please organize the dataset as follows:
The default storage path for training data is `PaddleOCR/train_data`, if you already have a dataset on your disk, just create a soft link to the dataset directory:
```
# linux and mac os
ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
# windows
mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
```
<a name="Costom_Dataset"></a>
### 1.1 Costom Dataset
If you want to use your own data for training, please refer to the following to organize your data.
- Training set
It is recommended to put the training images in the same folder, and use a txt file (rec_gt_train.txt) to store the image path and label. The contents of the txt file are as follows:
* Note: by default, the image path and image label are split with \t, if you use other methods to split, it will cause training error
```
" Image file name Image annotation "
train_data/rec/train/word_001.jpg 简单可依赖
train_data/rec/train/word_002.jpg 用科技让复杂的世界更简单
...
```
The final training set should have the following file structure:
```
|-train_data
|-rec
|- rec_gt_train.txt
|- train
|- word_001.png
|- word_002.jpg
|- word_003.jpg
| ...
```
- Test set
Similar to the training set, the test set also needs to be provided a folder containing all images (test) and a rec_gt_test.txt. The structure of the test set is as follows:
```
|-train_data
|-rec
|-ic15_data
|- rec_gt_test.txt
|- test
|- word_001.jpg
|- word_002.jpg
|- word_003.jpg
| ...
```
<a name="Dataset_download"></a>
### 1.2 Dataset Download
- ICDAR2015
If you do not have a dataset locally, you can download it on the official website [icdar2015](http://rrc.cvc.uab.es/?ch=4&com=downloads).
Also refer to [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) ,download the lmdb format dataset required for benchmark
If you want to reproduce the paper SAR, you need to download extra dataset [SynthAdd](https://pan.baidu.com/share/init?surl=uV0LtoNmcxbO-0YA7Ch4dg), extraction code: 627x. Besides, icdar2013, icdar2015, cocotext, IIIT5k datasets are also used to train. For specific details, please refer to the paper SAR. If you want to reproduce the paper SAR, you need to download extra dataset [SynthAdd](https://pan.baidu.com/share/init?surl=uV0LtoNmcxbO-0YA7Ch4dg), extraction code: 627x. Besides, icdar2013, icdar2015, cocotext, IIIT5k datasets are also used to train. For specific details, please refer to the paper SAR.
PaddleOCR provides label files for training the icdar2015 dataset, which can be downloaded in the following ways:
```
# Training set label
wget -P ./train_data/ic15_data https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt
# Test Set Label
wget -P ./train_data/ic15_data https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt
```
PaddleOCR also provides a data format conversion script, which can convert ICDAR official website label to a data format
supported by PaddleOCR. The data conversion tool is in `ppocr/utils/gen_label.py`, here is the training set as an example:
```
# convert the official gt to rec_gt_label.txt
python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
```
The data format is as follows, (a) is the original picture, (b) is the Ground Truth text file corresponding to each picture:
![](../datasets/icdar_rec.png)
- Multilingual dataset
The multi-language model training method is the same as the Chinese model. The training data set is 100w synthetic data. A small amount of fonts and test data can be downloaded using the following two methods.
* [Baidu Netdisk](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) ,Extraction code:frgi.
* [Google drive](https://drive.google.com/file/d/18cSWX7wXSy4G0tbKJ0d9PuIaiwRLHpjA/view)
<a name="Dictionary"></a> <a name="Dictionary"></a>
### 1.3 Dictionary ### 1.2 Dictionary
Finally, a dictionary ({word_dict_name}.txt) needs to be provided so that when the model is trained, all the characters that appear can be mapped to the dictionary index. Finally, a dictionary ({word_dict_name}.txt) needs to be provided so that when the model is trained, all the characters that appear can be mapped to the dictionary index.
......
...@@ -94,7 +94,7 @@ The current open source models, data sets and magnitudes are as follows: ...@@ -94,7 +94,7 @@ The current open source models, data sets and magnitudes are as follows:
- Chinese data set, LSVT street view data set crops the image according to the truth value, and performs position calibration, a total of 30w images. In addition, based on the LSVT corpus, 500w of synthesized data. - Chinese data set, LSVT street view data set crops the image according to the truth value, and performs position calibration, a total of 30w images. In addition, based on the LSVT corpus, 500w of synthesized data.
- Small language data set, using different corpora and fonts, respectively generated 100w synthetic data set, and using ICDAR-MLT as the verification set. - Small language data set, using different corpora and fonts, respectively generated 100w synthetic data set, and using ICDAR-MLT as the verification set.
Among them, the public data sets are all open source, users can search and download by themselves, or refer to [Chinese data set](./datasets_en.md), synthetic data is not open source, users can use open source synthesis tools to synthesize by themselves. Synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) etc. Among them, the public data sets are all open source, users can search and download by themselves, or refer to [Chinese data set](dataset/datasets_en.md), synthetic data is not open source, users can use open source synthesis tools to synthesize by themselves. Synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) etc.
<a name="22-vertical-scene"></a> <a name="22-vertical-scene"></a>
......
...@@ -19,7 +19,7 @@ ...@@ -19,7 +19,7 @@
- 2020.7.15, Add several related datasets, data annotation and synthesis tools. - 2020.7.15, Add several related datasets, data annotation and synthesis tools.
- 2020.7.9 Add a new model to support recognize the character "space". - 2020.7.9 Add a new model to support recognize the character "space".
- 2020.7.9 Add the data augument and learning rate decay strategies during training. - 2020.7.9 Add the data augument and learning rate decay strategies during training.
- 2020.6.8 Add [datasets](./datasets_en.md) and keep updating - 2020.6.8 Add [datasets](dataset/datasets_en.md) and keep updating
- 2020.6.5 Support exporting `attention` model to `inference_model` - 2020.6.5 Support exporting `attention` model to `inference_model`
- 2020.6.5 Support separate prediction and recognition, output result score - 2020.6.5 Support separate prediction and recognition, output result score
- 2020.5.30 Provide Lightweight Chinese OCR online experience - 2020.5.30 Provide Lightweight Chinese OCR online experience
......
...@@ -31,13 +31,14 @@ def get_bias_attr(k): ...@@ -31,13 +31,14 @@ def get_bias_attr(k):
class Head(nn.Layer): class Head(nn.Layer):
def __init__(self, in_channels, name_list): def __init__(self, in_channels, name_list, kernel_list=[3, 2, 2], **kwargs):
super(Head, self).__init__() super(Head, self).__init__()
self.conv1 = nn.Conv2D( self.conv1 = nn.Conv2D(
in_channels=in_channels, in_channels=in_channels,
out_channels=in_channels // 4, out_channels=in_channels // 4,
kernel_size=3, kernel_size=kernel_list[0],
padding=1, padding=int(kernel_list[0] // 2),
weight_attr=ParamAttr(), weight_attr=ParamAttr(),
bias_attr=False) bias_attr=False)
self.conv_bn1 = nn.BatchNorm( self.conv_bn1 = nn.BatchNorm(
...@@ -50,7 +51,7 @@ class Head(nn.Layer): ...@@ -50,7 +51,7 @@ class Head(nn.Layer):
self.conv2 = nn.Conv2DTranspose( self.conv2 = nn.Conv2DTranspose(
in_channels=in_channels // 4, in_channels=in_channels // 4,
out_channels=in_channels // 4, out_channels=in_channels // 4,
kernel_size=2, kernel_size=kernel_list[1],
stride=2, stride=2,
weight_attr=ParamAttr( weight_attr=ParamAttr(
initializer=paddle.nn.initializer.KaimingUniform()), initializer=paddle.nn.initializer.KaimingUniform()),
...@@ -65,7 +66,7 @@ class Head(nn.Layer): ...@@ -65,7 +66,7 @@ class Head(nn.Layer):
self.conv3 = nn.Conv2DTranspose( self.conv3 = nn.Conv2DTranspose(
in_channels=in_channels // 4, in_channels=in_channels // 4,
out_channels=1, out_channels=1,
kernel_size=2, kernel_size=kernel_list[2],
stride=2, stride=2,
weight_attr=ParamAttr( weight_attr=ParamAttr(
initializer=paddle.nn.initializer.KaimingUniform()), initializer=paddle.nn.initializer.KaimingUniform()),
...@@ -100,8 +101,8 @@ class DBHead(nn.Layer): ...@@ -100,8 +101,8 @@ class DBHead(nn.Layer):
'conv2d_57', 'batch_norm_49', 'conv2d_transpose_2', 'batch_norm_50', 'conv2d_57', 'batch_norm_49', 'conv2d_transpose_2', 'batch_norm_50',
'conv2d_transpose_3', 'thresh' 'conv2d_transpose_3', 'thresh'
] ]
self.binarize = Head(in_channels, binarize_name_list) self.binarize = Head(in_channels, binarize_name_list, **kwargs)
self.thresh = Head(in_channels, thresh_name_list) self.thresh = Head(in_channels, thresh_name_list, **kwargs)
def step_function(self, x, y): def step_function(self, x, y):
return paddle.reciprocal(1 + paddle.exp(-self.k * (x - y))) return paddle.reciprocal(1 + paddle.exp(-self.k * (x - y)))
......
...@@ -16,7 +16,7 @@ __all__ = ['build_neck'] ...@@ -16,7 +16,7 @@ __all__ = ['build_neck']
def build_neck(config): def build_neck(config):
from .db_fpn import DBFPN from .db_fpn import DBFPN, RSEFPN, LKPAN
from .east_fpn import EASTFPN from .east_fpn import EASTFPN
from .sast_fpn import SASTFPN from .sast_fpn import SASTFPN
from .rnn import SequenceEncoder from .rnn import SequenceEncoder
...@@ -26,8 +26,8 @@ def build_neck(config): ...@@ -26,8 +26,8 @@ def build_neck(config):
from .fce_fpn import FCEFPN from .fce_fpn import FCEFPN
from .pren_fpn import PRENFPN from .pren_fpn import PRENFPN
support_dict = [ support_dict = [
'FPN', 'FCEFPN', 'DBFPN', 'EASTFPN', 'SASTFPN', 'SequenceEncoder', 'FPN', 'FCEFPN', 'LKPAN', 'DBFPN', 'RSEFPN', 'EASTFPN', 'SASTFPN',
'PGFPN', 'TableFPN', 'PRENFPN' 'SequenceEncoder', 'PGFPN', 'TableFPN', 'PRENFPN'
] ]
module_name = config.pop('name') module_name = config.pop('name')
......
...@@ -20,6 +20,88 @@ import paddle ...@@ -20,6 +20,88 @@ import paddle
from paddle import nn from paddle import nn
import paddle.nn.functional as F import paddle.nn.functional as F
from paddle import ParamAttr from paddle import ParamAttr
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(__dir__)
sys.path.insert(0, os.path.abspath(os.path.join(__dir__, '../../..')))
from ppocr.modeling.backbones.det_mobilenet_v3 import SEModule
class DSConv(nn.Layer):
def __init__(self,
in_channels,
out_channels,
kernel_size,
padding,
stride=1,
groups=None,
if_act=True,
act="relu",
**kwargs):
super(DSConv, self).__init__()
if groups == None:
groups = in_channels
self.if_act = if_act
self.act = act
self.conv1 = nn.Conv2D(
in_channels=in_channels,
out_channels=in_channels,
kernel_size=kernel_size,
stride=stride,
padding=padding,
groups=groups,
bias_attr=False)
self.bn1 = nn.BatchNorm(num_channels=in_channels, act=None)
self.conv2 = nn.Conv2D(
in_channels=in_channels,
out_channels=int(in_channels * 4),
kernel_size=1,
stride=1,
bias_attr=False)
self.bn2 = nn.BatchNorm(num_channels=int(in_channels * 4), act=None)
self.conv3 = nn.Conv2D(
in_channels=int(in_channels * 4),
out_channels=out_channels,
kernel_size=1,
stride=1,
bias_attr=False)
self._c = [in_channels, out_channels]
if in_channels != out_channels:
self.conv_end = nn.Conv2D(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
stride=1,
bias_attr=False)
def forward(self, inputs):
x = self.conv1(inputs)
x = self.bn1(x)
x = self.conv2(x)
x = self.bn2(x)
if self.if_act:
if self.act == "relu":
x = F.relu(x)
elif self.act == "hardswish":
x = F.hardswish(x)
else:
print("The activation function({}) is selected incorrectly.".
format(self.act))
exit()
x = self.conv3(x)
if self._c[0] != self._c[1]:
x = x + self.conv_end(inputs)
return x
class DBFPN(nn.Layer): class DBFPN(nn.Layer):
...@@ -106,3 +188,171 @@ class DBFPN(nn.Layer): ...@@ -106,3 +188,171 @@ class DBFPN(nn.Layer):
fuse = paddle.concat([p5, p4, p3, p2], axis=1) fuse = paddle.concat([p5, p4, p3, p2], axis=1)
return fuse return fuse
class RSELayer(nn.Layer):
def __init__(self, in_channels, out_channels, kernel_size, shortcut=True):
super(RSELayer, self).__init__()
weight_attr = paddle.nn.initializer.KaimingUniform()
self.out_channels = out_channels
self.in_conv = nn.Conv2D(
in_channels=in_channels,
out_channels=self.out_channels,
kernel_size=kernel_size,
padding=int(kernel_size // 2),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.se_block = SEModule(self.out_channels)
self.shortcut = shortcut
def forward(self, ins):
x = self.in_conv(ins)
if self.shortcut:
out = x + self.se_block(x)
else:
out = self.se_block(x)
return out
class RSEFPN(nn.Layer):
def __init__(self, in_channels, out_channels, shortcut=True, **kwargs):
super(RSEFPN, self).__init__()
self.out_channels = out_channels
self.ins_conv = nn.LayerList()
self.inp_conv = nn.LayerList()
for i in range(len(in_channels)):
self.ins_conv.append(
RSELayer(
in_channels[i],
out_channels,
kernel_size=1,
shortcut=shortcut))
self.inp_conv.append(
RSELayer(
out_channels,
out_channels // 4,
kernel_size=3,
shortcut=shortcut))
def forward(self, x):
c2, c3, c4, c5 = x
in5 = self.ins_conv[3](c5)
in4 = self.ins_conv[2](c4)
in3 = self.ins_conv[1](c3)
in2 = self.ins_conv[0](c2)
out4 = in4 + F.upsample(
in5, scale_factor=2, mode="nearest", align_mode=1) # 1/16
out3 = in3 + F.upsample(
out4, scale_factor=2, mode="nearest", align_mode=1) # 1/8
out2 = in2 + F.upsample(
out3, scale_factor=2, mode="nearest", align_mode=1) # 1/4
p5 = self.inp_conv[3](in5)
p4 = self.inp_conv[2](out4)
p3 = self.inp_conv[1](out3)
p2 = self.inp_conv[0](out2)
p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
fuse = paddle.concat([p5, p4, p3, p2], axis=1)
return fuse
class LKPAN(nn.Layer):
def __init__(self, in_channels, out_channels, mode='large', **kwargs):
super(LKPAN, self).__init__()
self.out_channels = out_channels
weight_attr = paddle.nn.initializer.KaimingUniform()
self.ins_conv = nn.LayerList()
self.inp_conv = nn.LayerList()
# pan head
self.pan_head_conv = nn.LayerList()
self.pan_lat_conv = nn.LayerList()
if mode.lower() == 'lite':
p_layer = DSConv
elif mode.lower() == 'large':
p_layer = nn.Conv2D
else:
raise ValueError(
"mode can only be one of ['lite', 'large'], but received {}".
format(mode))
for i in range(len(in_channels)):
self.ins_conv.append(
nn.Conv2D(
in_channels=in_channels[i],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False))
self.inp_conv.append(
p_layer(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=9,
padding=4,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False))
if i > 0:
self.pan_head_conv.append(
nn.Conv2D(
in_channels=self.out_channels // 4,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
stride=2,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False))
self.pan_lat_conv.append(
p_layer(
in_channels=self.out_channels // 4,
out_channels=self.out_channels // 4,
kernel_size=9,
padding=4,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False))
def forward(self, x):
c2, c3, c4, c5 = x
in5 = self.ins_conv[3](c5)
in4 = self.ins_conv[2](c4)
in3 = self.ins_conv[1](c3)
in2 = self.ins_conv[0](c2)
out4 = in4 + F.upsample(
in5, scale_factor=2, mode="nearest", align_mode=1) # 1/16
out3 = in3 + F.upsample(
out4, scale_factor=2, mode="nearest", align_mode=1) # 1/8
out2 = in2 + F.upsample(
out3, scale_factor=2, mode="nearest", align_mode=1) # 1/4
f5 = self.inp_conv[3](in5)
f4 = self.inp_conv[2](out4)
f3 = self.inp_conv[1](out3)
f2 = self.inp_conv[0](out2)
pan3 = f3 + self.pan_head_conv[0](f2)
pan4 = f4 + self.pan_head_conv[1](pan3)
pan5 = f5 + self.pan_head_conv[2](pan4)
p2 = self.pan_lat_conv[0](f2)
p3 = self.pan_lat_conv[1](pan3)
p4 = self.pan_lat_conv[2](pan4)
p5 = self.pan_lat_conv[3](pan5)
p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
fuse = paddle.concat([p5, p4, p3, p2], axis=1)
return fuse
...@@ -158,7 +158,7 @@ class TextDetector(object): ...@@ -158,7 +158,7 @@ class TextDetector(object):
rect[1] = pts[np.argmin(diff)] rect[1] = pts[np.argmin(diff)]
rect[3] = pts[np.argmax(diff)] rect[3] = pts[np.argmax(diff)]
return rect return rect
def clip_det_res(self, points, img_height, img_width): def clip_det_res(self, points, img_height, img_width):
for pno in range(points.shape[0]): for pno in range(points.shape[0]):
points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1)) points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
...@@ -284,7 +284,7 @@ if __name__ == "__main__": ...@@ -284,7 +284,7 @@ if __name__ == "__main__":
total_time += elapse total_time += elapse
count += 1 count += 1
save_pred = os.path.basename(image_file) + "\t" + str( save_pred = os.path.basename(image_file) + "\t" + str(
json.dumps(np.array(dt_boxes).astype(np.int32).tolist())) + "\n" json.dumps([x.tolist() for x in dt_boxes])) + "\n"
save_results.append(save_pred) save_results.append(save_pred)
logger.info(save_pred) logger.info(save_pred)
logger.info("The predict time of {}: {}".format(image_file, elapse)) logger.info("The predict time of {}: {}".format(image_file, elapse))
......
Markdown is supported
0% .
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