未验证 提交 2b6c887a 编写于 作者: M MissPenguin 提交者: GitHub

Merge pull request #6074 from MissPenguin/dygraph

update docs
...@@ -22,9 +22,11 @@ PP-OCR has supported muti deployment schemes. Click the link to get the specific ...@@ -22,9 +22,11 @@ PP-OCR has supported muti deployment schemes. Click the link to get the specific
- [Python Inference](../doc/doc_en/inference_ppocr_en.md) - [Python Inference](../doc/doc_en/inference_ppocr_en.md)
- [C++ Inference](./cpp_infer/readme.md) - [C++ Inference](./cpp_infer/readme.md)
- [Serving](./pdserving/README.md) - [Serving (Python/C++)](./pdserving/README.md)
- [Paddle-Lite](./lite/readme.md) - [Paddle-Lite (ARM CPU/OpenCL ARM GPU/Metal ARM GPU)](./lite/readme.md)
- [Paddle.js](./paddlejs/README.md) - [Paddle.js](./paddlejs/README.md)
- [Jetson Inference]()
- [XPU Inference]()
- [Paddle2ONNX](./paddle2onnx/readme.md) - [Paddle2ONNX](./paddle2onnx/readme.md)
If you need the deployment tutorial of academic algorithm models other than PP-OCR, please directly enter the main page of corresponding algorithms, [entrance](../doc/doc_en/algorithm_overview_en.md) If you need the deployment tutorial of academic algorithm models other than PP-OCR, please directly enter the main page of corresponding algorithms, [entrance](../doc/doc_en/algorithm_overview_en.md)
\ No newline at end of file
...@@ -22,9 +22,11 @@ PP-OCR模型已打通多种场景部署方案,点击链接获取具体的使 ...@@ -22,9 +22,11 @@ PP-OCR模型已打通多种场景部署方案,点击链接获取具体的使
- [Python 推理](../doc/doc_ch/inference_ppocr.md) - [Python 推理](../doc/doc_ch/inference_ppocr.md)
- [C++ 推理](./cpp_infer/readme_ch.md) - [C++ 推理](./cpp_infer/readme_ch.md)
- [Serving 服务化部署](./pdserving/README_CN.md) - [Serving 服务化部署(Python/C++)](./pdserving/README_CN.md)
- [Paddle-Lite 端侧部署](./lite/readme_ch.md) - [Paddle-Lite 端侧部署(ARM CPU/OpenCL ARM GPU/Metal ARM GPU)](./lite/readme_ch.md)
- [Paddle.js 服务化部署](./paddlejs/README_ch.md) - [Paddle.js 部署](./paddlejs/README_ch.md)
- [Jetson 推理]()
- [XPU 推理]()
- [Paddle2ONNX 推理](./paddle2onnx/readme_ch.md) - [Paddle2ONNX 推理](./paddle2onnx/readme_ch.md)
需要PP-OCR以外的学术算法模型的推理部署,请直接进入相应算法主页面,[入口](../doc/doc_ch/algorithm_overview.md) 需要PP-OCR以外的学术算法模型的推理部署,请直接进入相应算法主页面,[入口](../doc/doc_ch/algorithm_overview.md)
\ No newline at end of file
...@@ -25,8 +25,8 @@ ...@@ -25,8 +25,8 @@
|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接| |模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
| --- | --- | --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- | --- | --- |
|DB|ResNet50_vd|configs/det/det_r50_vd_db.yml|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)| |DB|ResNet50_vd|[configs/det/det_r50_vd_db.yml](../../configs/det/det_r50_vd_db.yml)|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
|DB|MobileNetV3|configs/det/det_mv3_db.yml|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)| |DB|MobileNetV3|[configs/det/det_mv3_db.yml](../../configs/det/det_mv3_db.yml)|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
<a name="2"></a> <a name="2"></a>
......
...@@ -17,6 +17,8 @@ ...@@ -17,6 +17,8 @@
PP-OCR是PaddleOCR自研的实用的超轻量OCR系统。在实现[前沿算法](algorithm.md)的基础上,考虑精度与速度的平衡,进行**模型瘦身****深度优化**,使其尽可能满足产业落地需求。 PP-OCR是PaddleOCR自研的实用的超轻量OCR系统。在实现[前沿算法](algorithm.md)的基础上,考虑精度与速度的平衡,进行**模型瘦身****深度优化**,使其尽可能满足产业落地需求。
#### PP-OCR
PP-OCR是一个两阶段的OCR系统,其中文本检测算法选用[DB](algorithm_det_db.md),文本识别算法选用[CRNN](algorithm_rec_crnn.md),并在检测和识别模块之间添加[文本方向分类器](angle_class.md),以应对不同方向的文本识别。 PP-OCR是一个两阶段的OCR系统,其中文本检测算法选用[DB](algorithm_det_db.md),文本识别算法选用[CRNN](algorithm_rec_crnn.md),并在检测和识别模块之间添加[文本方向分类器](angle_class.md),以应对不同方向的文本识别。
PP-OCR系统pipeline如下: PP-OCR系统pipeline如下:
...@@ -28,9 +30,13 @@ PP-OCR系统pipeline如下: ...@@ -28,9 +30,13 @@ PP-OCR系统pipeline如下:
PP-OCR系统在持续迭代优化,目前已发布PP-OCR和PP-OCRv2两个版本: PP-OCR系统在持续迭代优化,目前已发布PP-OCR和PP-OCRv2两个版本:
[1] PP-OCR从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面,采用19个有效策略,对各个模块的模型进行效果调优和瘦身(如绿框所示),最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941 PP-OCR从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面,采用19个有效策略,对各个模块的模型进行效果调优和瘦身(如绿框所示),最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941
#### PP-OCRv2
PP-OCRv2在PP-OCR的基础上,进一步在5个方面重点优化,检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略;识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](./doc/doc_ch/enhanced_ctc_loss.md)损失函数改进(如上图红框所示),进一步在推理速度和预测效果上取得明显提升。更多细节请参考PP-OCRv2[技术报告](https://arxiv.org/abs/2109.03144)
[2] PP-OCRv2在PP-OCR的基础上,进一步在5个方面重点优化,检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略;识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](./doc/doc_ch/enhanced_ctc_loss.md)损失函数改进(如上图红框所示),进一步在推理速度和预测效果上取得明显提升。更多细节请参考PP-OCRv2[技术报告](https://arxiv.org/abs/2109.03144) #### PP-OCRv3
<a name="2"></a> <a name="2"></a>
......
...@@ -25,8 +25,8 @@ On the ICDAR2015 dataset, the text detection result is as follows: ...@@ -25,8 +25,8 @@ On the ICDAR2015 dataset, the text detection result is as follows:
|Model|Backbone|Configuration|Precision|Recall|Hmean|Download| |Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
| --- | --- | --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- | --- | --- |
|DB|ResNet50_vd|configs/det/det_r50_vd_db.yml|86.41%|78.72%|82.38%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)| |DB|ResNet50_vd|[configs/det/det_r50_vd_db.yml](../../configs/det/det_r50_vd_db.yml)|86.41%|78.72%|82.38%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
|DB|MobileNetV3|configs/det/det_mv3_db.yml|77.29%|73.08%|75.12%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)| |DB|MobileNetV3|[configs/det/det_mv3_db.yml](../../configs/det/det_mv3_db.yml)|77.29%|73.08%|75.12%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
<a name="2"></a> <a name="2"></a>
......
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