Merge remote-tracking branch 'upstream/dygraph' into dy1

configs/rec/rec_mv3_tps_bilstm_att.yml

+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/rec_mv3_tps_bilstm_att/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: 
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: RARE
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: small
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn 
+    hidden_size: 96
+  Head:
+    name: AttentionHead  
+    hidden_size: 96
+    
+
+Loss:
+  name: AttentionLoss
+
+PostProcess:
+  name: AttnLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 1

configs/rec/rec_mv3_tps_bilstm_ctc.yml

 Global:
-  use_gpu: true
+  use_gpu: True
   epoch_num: 72
   log_smooth_window: 20
   print_batch_step: 10
@@ -7,7 +7,6 @@ Global:
   save_epoch_step: 3
   # evaluation is run every 2000 iterations
   eval_batch_step: [0, 2000]
-  # if pretrained_model is saved in static mode, load_static_weights must set to True
   cal_metric_during_train: True
   pretrained_model:
   checkpoints:

configs/rec/rec_r34_vd_none_bilstm_ctc.yml

@@ -7,7 +7,6 @@ Global:
   save_epoch_step: 3
   # evaluation is run every 2000 iterations
   eval_batch_step: [0, 2000]
-  # if pretrained_model is saved in static mode, load_static_weights must set to True
   cal_metric_during_train: True
   pretrained_model:
   checkpoints:

configs/rec/rec_r34_vd_none_none_ctc.yml

@@ -7,7 +7,6 @@ Global:
   save_epoch_step: 3
   # evaluation is run every 2000 iterations
   eval_batch_step: [0, 2000]
-  # if pretrained_model is saved in static mode, load_static_weights must set to True
   cal_metric_during_train: True
   pretrained_model:
   checkpoints:

configs/rec/rec_r34_vd_tps_bilstm_att.yml

+Global:
+  use_gpu: True
+  epoch_num: 400
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/b3_rare_r34_none_gru/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: 
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: rec
+  algorithm: RARE
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: large
+  Backbone:
+    name: ResNet  
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn 
+    hidden_size: 256 #96
+  Head:
+    name: AttentionHead  # AttentionHead
+    hidden_size: 256 #
+    l2_decay: 0.00001
+
+Loss:
+  name: AttentionLoss
+
+PostProcess:
+  name: AttnLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8

configs/rec/rec_r34_vd_tps_bilstm_ctc.yml

@@ -7,7 +7,6 @@ Global:
   save_epoch_step: 3
   # evaluation is run every 2000 iterations
   eval_batch_step: [0, 2000]
-  # if pretrained_model is saved in static mode, load_static_weights must set to True
   cal_metric_during_train: True
   pretrained_model:
   checkpoints:

configs/rec/rec_r50_fpn_srn.yml

@@ -7,7 +7,6 @@ Global:
   save_epoch_step: 3
   # evaluation is run every 5000 iterations after the 4000th iteration
   eval_batch_step: [0, 5000]
-  # if pretrained_model is saved in static mode, load_static_weights must set to True
   cal_metric_during_train: True
   pretrained_model: 
   checkpoints:
@@ -59,7 +58,7 @@ Metric:
 Train:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/srn_train_data_duiqi
+    data_dir: ./train_data/data_lmdb_release/training/
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
@@ -84,7 +83,7 @@ Train:
 Eval:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/evaluation
+    data_dir: ./train_data/data_lmdb_release/validation/
     transforms:
       - DecodeImage: # load image
           img_mode: BGR

deploy/cpp_infer/CMakeLists.txt

@@ -133,7 +133,11 @@ if(WITH_MKL)
     endif ()
   endif()
 else()
+  if (WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/openblas${CMAKE_STATIC_LIBRARY_SUFFIX})
+  else ()
     set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/libopenblas${CMAKE_STATIC_LIBRARY_SUFFIX})
+  endif ()
 endif()
 
 # Note: libpaddle_inference_api.so/a must put before libpaddle_fluid.so/a

deploy/cpp_infer/docs/windows_vs2019_build.md

@@ -14,7 +14,7 @@ PaddleOCR在Windows 平台下基于`Visual Studio 2019 Community` 进行了测
 
 ### Step1: 下载PaddlePaddle C++ 预测库 fluid_inference
 
-PaddlePaddle C++ 预测库针对不同的`CPU`和`CUDA`版本提供了不同的预编译版本，请根据实际情况下载:  [C++预测库下载列表](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/advanced_guide/inference_deployment/inference/windows_cpp_inference.html)
+PaddlePaddle C++ 预测库针对不同的`CPU`和`CUDA`版本提供了不同的预编译版本，请根据实际情况下载:  [C++预测库下载列表](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/windows_cpp_inference.html)
 
 解压后`D:\projects\fluid_inference`目录包含内容为：
 ```

deploy/cpp_infer/readme.md

 # 服务器端C++预测
 
-本教程将介绍在服务器端部署PaddleOCR超轻量中文检测、识别模型的详细步骤。
+本章节介绍PaddleOCR 模型的的C++部署方法，与之对应的python预测部署方式参考[文档](../../doc/doc_ch/inference.md)。
+C++在性能计算上优于python，因此，在大多数CPU、GPU部署场景，多采用C++的部署方式，本节将介绍如何在Linux\Windows (CPU\GPU)环境下配置C++环境并完成
+PaddleOCR模型部署。
 
 
 ## 1. 准备环境
@@ -72,9 +74,21 @@ opencv3/
 
 * 有2种方式获取Paddle预测库，下面进行详细介绍。
 
-#### 1.2.1 预测库源码编译
+#### 1.2.1 直接下载安装
+
+* [Paddle预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html)上提供了不同cuda版本的Linux预测库，可以在官网查看并选择合适的预测库版本。
+
+* 下载之后使用下面的方法解压。
+
+```
+tar -xf paddle_inference.tgz
+```
+
+最终会在当前的文件夹中生成`paddle_inference/`的子文件夹。
+
+#### 1.2.2 预测库源码编译
 * 如果希望获取最新预测库特性，可以从Paddle github上克隆最新代码，源码编译预测库。
-* 可以参考[Paddle预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/advanced_guide/inference_deployment/inference/build_and_install_lib_cn.html)的说明，从github上获取Paddle代码，然后进行编译，生成最新的预测库。使用git获取代码方法如下。
+* 可以参考[Paddle预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html)的说明，从github上获取Paddle代码，然后进行编译，生成最新的预测库。使用git获取代码方法如下。
 
 ```shell
 git clone https://github.com/PaddlePaddle/Paddle.git
@@ -100,7 +114,7 @@ make -j
 make inference_lib_dist
 ```
 
-更多编译参数选项可以参考Paddle C++预测库官网：[https://www.paddlepaddle.org.cn/documentation/docs/zh/advanced_guide/inference_deployment/inference/build_and_install_lib_cn.html](https://www.paddlepaddle.org.cn/documentation/docs/zh/advanced_guide/inference_deployment/inference/build_and_install_lib_cn.html)。
+更多编译参数选项可以参考Paddle C++预测库官网：[https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html)。
 
 
 * 编译完成之后，可以在`build/paddle_inference_install_dir/`文件下看到生成了以下文件及文件夹。
@@ -115,17 +129,7 @@ build/paddle_inference_install_dir/
 
 其中`paddle`就是C++预测所需的Paddle库，`version.txt`中包含当前预测库的版本信息。
 
-#### 1.2.2 直接下载安装
-
-* [Paddle预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/advanced_guide/inference_deployment/inference/build_and_install_lib_cn.html)上提供了不同cuda版本的Linux预测库，可以在官网查看并选择合适的预测库版本。
 
-* 下载之后使用下面的方法解压。
-
-```
-tar -xf paddle_inference.tgz
-```
-
-最终会在当前的文件夹中生成`paddle_inference/`的子文件夹。
 
 
 ## 2 开始运行
@@ -223,7 +227,7 @@ char_list_file ../../ppocr/utils/ppocr_keys_v1.txt # 字典文件
 visualize 1 # 是否对结果进行可视化，为1时，会在当前文件夹下保存文件名为`ocr_vis.png`的预测结果。
 ```
 
-* PaddleOCR也支持多语言的预测，更多细节可以参考[识别文档](../../doc/doc_ch/recognition.md)中的多语言字典与模型部分。
+* PaddleOCR也支持多语言的预测，更多支持的语言和模型可以参考[识别文档](../../doc/doc_ch/recognition.md)中的多语言字典与模型部分，如果希望进行多语言预测，只需将修改`tools/config.txt`中的`char_list_file`（字典文件路径）以及`rec_model_dir`（inference模型路径）字段即可。
 
 最终屏幕上会输出检测结果如下。
 
@@ -234,4 +238,4 @@ visualize 1 # 是否对结果进行可视化，为1时，会在当前文件夹
 
 ### 2.3 注意
 
-* 在使用Paddle预测库时，推荐使用2.0.0-beta0版本的预测库。
+* 在使用Paddle预测库时，推荐使用2.0.0版本的预测库。

deploy/cpp_infer/readme_en.md

 # Server-side C++ inference
 
-
-In this tutorial, we will introduce the detailed steps of deploying PaddleOCR ultra-lightweight Chinese detection and recognition models on the server side.
+This chapter introduces the C++ deployment method of the PaddleOCR model, and the corresponding python predictive deployment method refers to [document](../../doc/doc_ch/inference.md).
+C++ is better than python in terms of performance calculation. Therefore, in most CPU and GPU deployment scenarios, C++ deployment is mostly used.
+This section will introduce how to configure the C++ environment and complete it in the Linux\Windows (CPU\GPU) environment
+PaddleOCR model deployment.
 
 
 ## 1. Prepare the environment
@@ -74,10 +76,23 @@ opencv3/
 
 * There are 2 ways to obtain the Paddle inference library, described in detail below.
 
+#### 1.2.1 Direct download and installation
+
+* Different cuda versions of the Linux inference library (based on GCC 4.8.2) are provided on the
+[Paddle inference library official website](https://www.paddlepaddle.org.cn/documentation/docs/en/develop/guides/05_inference_deployment/inference/build_and_install_lib_en.html). You can view and select the appropriate version of the inference library on the official website.
+
 
-#### 1.2.1 Compile from the source code
+* After downloading, use the following method to uncompress.
+
+```
+tar -xf paddle_inference.tgz
+```
+
+Finally you can see the following files in the folder of `paddle_inference/`.
+
+#### 1.2.2 Compile from the source code
 * If you want to get the latest Paddle inference library features, you can download the latest code from Paddle github repository and compile the inference library from the source code.
-* You can refer to [Paddle inference library] (https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html) to get the Paddle source code from github, and then compile To generate the latest inference library. The method of using git to access the code is as follows.
+* You can refer to [Paddle inference library] (https://www.paddlepaddle.org.cn/documentation/docs/en/develop/guides/05_inference_deployment/inference/build_and_install_lib_en.html) to get the Paddle source code from github, and then compile To generate the latest inference library. The method of using git to access the code is as follows.
 
 
 ```shell
@@ -104,7 +119,7 @@ make -j
 make inference_lib_dist
 ```
 
-For more compilation parameter options, please refer to the official website of the Paddle C++ inference library:[https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html](https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html).
+For more compilation parameter options, please refer to the official website of the Paddle C++ inference library:[https://www.paddlepaddle.org.cn/documentation/docs/en/develop/guides/05_inference_deployment/inference/build_and_install_lib_en.html](https://www.paddlepaddle.org.cn/documentation/docs/en/develop/guides/05_inference_deployment/inference/build_and_install_lib_en.html).
 
 
 * After the compilation process, you can see the following files in the folder of `build/paddle_inference_install_dir/`.
@@ -120,22 +135,6 @@ build/paddle_inference_install_dir/
 Among them, `paddle` is the Paddle library required for C++ prediction later, and `version.txt` contains the version information of the current inference library.
 
 
-
-#### 1.2.2 Direct download and installation
-
-* Different cuda versions of the Linux inference library (based on GCC 4.8.2) are provided on the
-[Paddle inference library official website](https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html). You can view and select the appropriate version of the inference library on the official website.
-
-
-* After downloading, use the following method to uncompress.
-
-```
-tar -xf paddle_inference.tgz
-```
-
-Finally you can see the following files in the folder of `paddle_inference/`.
-
-
 ## 2. Compile and run the demo
 
 ### 2.1 Export the inference model
@@ -233,7 +232,7 @@ char_list_file ../../ppocr/utils/ppocr_keys_v1.txt # dictionary file
 visualize 1 # Whether to visualize the results，when it is set as 1, The prediction result will be save in the image file `./ocr_vis.png`.
 ```
 
-* Multi-language inference is also supported in PaddleOCR, for more details, please refer to part of multi-language dictionaries and models in [recognition tutorial](../../doc/doc_en/recognition_en.md).
+* Multi-language inference is also supported in PaddleOCR, you can refer to [recognition tutorial](../../doc/doc_en/recognition_en.md) for more supported languages and models in PaddleOCR. Specifically, if you want to infer using multi-language models, you just need to modify values of `char_list_file` and `rec_model_dir` in file `tools/config.txt`.
 
 
 The detection results will be shown on the screen, which is as follows.
@@ -245,4 +244,4 @@ The detection results will be shown on the screen, which is as follows.
 
 ### 2.3 Notes
 
-* Paddle2.0.0-beta0 inference model library is recommended for this toturial.
+* Paddle2.0.0 inference model library is recommended for this toturial.

deploy/cpp_infer/tools/config.txt

@@ -9,7 +9,7 @@ use_mkldnn 0
 max_side_len  960
 det_db_thresh  0.3
 det_db_box_thresh  0.5
-det_db_unclip_ratio  2.0
+det_db_unclip_ratio  1.6
 det_model_dir  ./inference/ch_ppocr_mobile_v2.0_det_infer/
 
 # cls config

deploy/docker/hubserving/cpu/Dockerfile

 # Version: 2.0.0
-FROM registry.baidubce.com/paddlepaddle/paddle:2.0.0rc1
+FROM registry.baidubce.com/paddlepaddle/paddle:2.0.0
 
 # PaddleOCR base on Python3.7
 RUN pip3.7 install --upgrade pip -i https://mirror.baidu.com/pypi/simple

deploy/docker/hubserving/gpu/Dockerfile

 # Version: 2.0.0
-FROM egistry.baidubce.com/paddlepaddle/paddle:2.0.0rc1-gpu-cuda10.0-cudnn7
+FROM registry.baidubce.com/paddlepaddle/paddle:2.0.0-gpu-cuda10.1-cudnn7
 
 # PaddleOCR base on Python3.7
 RUN pip3.7 install --upgrade pip -i https://mirror.baidu.com/pypi/simple

deploy/hubserving/ocr_det/params.py

@@ -20,7 +20,8 @@ def read_params():
     #DB parmas
     cfg.det_db_thresh = 0.3
     cfg.det_db_box_thresh = 0.5
-    cfg.det_db_unclip_ratio = 2.0
+    cfg.det_db_unclip_ratio = 1.6
+    cfg.use_dilation = False
 
     # #EAST parmas
     # cfg.det_east_score_thresh = 0.8

deploy/hubserving/ocr_system/params.py

@@ -20,7 +20,8 @@ def read_params():
     #DB parmas
     cfg.det_db_thresh = 0.3
     cfg.det_db_box_thresh = 0.5
-    cfg.det_db_unclip_ratio = 2.0
+    cfg.det_db_unclip_ratio = 1.6
+    cfg.use_dilation = False
 
     #EAST parmas
     cfg.det_east_score_thresh = 0.8

deploy/slim/prune/README.md

+
+## 介绍
+
+复杂的模型有利于提高模型的性能，但也导致模型中存在一定冗余，模型裁剪通过移出网络模型中的子模型来减少这种冗余，达到减少模型计算复杂度，提高模型推理性能的目的。
+本教程将介绍如何使用飞桨模型压缩库PaddleSlim做PaddleOCR模型的压缩。
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)集成了模型剪枝、量化（包括量化训练和离线量化）、蒸馏和神经网络搜索等多种业界常用且领先的模型压缩功能，如果您感兴趣，可以关注并了解。
+
+
+在开始本教程之前，建议先了解：
+1. [PaddleOCR模型的训练方法](../../../doc/doc_ch/quickstart.md)
+2. [模型裁剪教程](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+
+## 快速开始
+
+模型裁剪主要包括四个步骤：
+1. 安装 PaddleSlim
+2. 准备训练好的模型
+3. 敏感度分析、裁剪训练
+4. 导出模型、预测部署
+
+### 1. 安装PaddleSlim
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+git checkout develop
+cd Paddleslim
+python3 setup.py install
+```
+
+### 2. 获取预训练模型
+模型裁剪需要加载事先训练好的模型，PaddleOCR也提供了一系列(模型)[../../../doc/doc_ch/models_list.md]，开发者可根据需要自行选择模型或使用自己的模型。
+
+### 3. 敏感度分析训练
+
+加载预训练模型后，通过对现有模型的每个网络层进行敏感度分析，得到敏感度文件：sen.pickle，可以通过PaddleSlim提供的[接口](https://github.com/PaddlePaddle/PaddleSlim/blob/9b01b195f0c4bc34a1ab434751cb260e13d64d9e/paddleslim/dygraph/prune/filter_pruner.py#L75)加载文件，获得各网络层在不同裁剪比例下的精度损失。从而了解各网络层冗余度，决定每个网络层的裁剪比例。
+敏感度文件内容格式：
+    sen.pickle(Dict){
+            'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+            'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+        }
+
+    例子：
+        {
+            'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+            'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+        }
+加载敏感度文件后会返回一个字典，字典中的keys为网络模型参数模型的名字，values为一个字典，里面保存了相应网络层的裁剪敏感度信息。例如在例子中，conv10_expand_weights所对应的网络层在裁掉10%的卷积核后模型性能相较原模型会下降0.65%，详细信息可见[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+进入PaddleOCR根目录，通过以下命令对模型进行敏感度分析训练：
+```bash
+python3.7 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights="your trained model"
+```
+
+### 4. 导出模型、预测部署
+
+在得到裁剪训练保存的模型后，我们可以将其导出为inference_model：
+```bash
+pytho3.7 deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights=./output/det_db/best_accuracy  Global.save_inference_dir=inference_model
+```
+
+inference model的预测和部署参考：
+1. [inference model python端预测](../../../doc/doc_ch/inference.md)
+2. [inference model C++预测](../../cpp_infer/readme.md)

deploy/slim/prune/README_en.md

+
+## Introduction
+
+Generally, a more complex model would achive better performance in the task, but it also leads to some redundancy in the model. Model Pruning is a technique that reduces this redundancy by removing the sub-models in the neural network model, so as to reduce model calculation complexity and improve model inference performance.
+
+This example uses PaddleSlim provided[APIs of Pruning](https://paddlepaddle.github.io/PaddleSlim/api/prune_api/) to compress the OCR model.
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim), an open source library which integrates model pruning, quantization (including quantization training and offline quantization), distillation, neural network architecture search, and many other commonly used and leading model compression technique in the industry.
+
+It is recommended that you could understand following pages before reading this example：
+1. [PaddleOCR training methods](../../../doc/doc_ch/quickstart.md)
+2. [The demo of prune](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+## Quick start
+
+Five steps for OCR model prune:
+1. Install PaddleSlim
+2. Prepare the trained model
+3. Sensitivity analysis and tailoring training
+4. Export model, predict deployment
+
+### 1. Install PaddleSlim
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+git checkout develop
+cd Paddleslim
+python3 setup.py install
+```
+
+
+### 2. Download Pretrain Model
+Model prune needs to load pre-trained models.
+PaddleOCR also provides a series of (models)[../../../doc/doc_en/models_list_en.md]. Developers can choose their own models or use their own models according to their needs.
+
+
+### 3. Pruning sensitivity analysis
+
+  After the pre-training model is loaded, sensitivity analysis is performed on each network layer of the model to understand the redundancy of each network layer, and save a sensitivity file which named: sen.pickle.  After that, user could load the sensitivity file via the [methods provided by PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/prune/sensitive.py#L221) and determining the pruning ratio of each network layer automatically. For specific details of sensitivity analysis, see：[Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/tutorials/image_classification_sensitivity_analysis_tutorial.md)
+  The data format of sensitivity file：
+      sen.pickle(Dict){
+              'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+              'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+          }
+
+      example：
+          {
+              'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+              'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+          }
+  The function would return a dict after loading the sensitivity file. The keys of the dict are name of parameters in each layer. And the value of key is the information about pruning sensitivity of correspoding layer. In example, pruning 10% filter of the layer corresponding to conv10_expand_weights would lead to 0.65% degradation of model performance. The details could be seen at: [Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+
+Enter the PaddleOCR root directory，perform sensitivity analysis on the model with the following command：
+
+```bash
+
+python3.7 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights="your trained model"
+
+```
+
+
+### 5.  Export inference model and deploy it
+
+We can export the pruned model as inference_model for deployment:
+```bash
+python deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml  -o Global.pretrain_weights=./output/det_db/best_accuracy Global.test_batch_size_per_card=1 Global.save_inference_dir=inference_model
+```
+
+Reference for prediction and deployment of inference model:
+1. [inference model python prediction](../../../doc/doc_en/inference_en.md)
+2. [inference model C++ prediction](../../cpp_infer/readme_en.md)

deploy/slim/prune/export_prune_model.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, '..', '..', '..'))
+sys.path.append(os.path.join(__dir__, '..', '..', '..', 'tools'))
+
+import paddle
+from ppocr.data import build_dataloader
+from ppocr.modeling.architectures import build_model
+
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import init_model
+import tools.program as program
+
+
+def main(config, device, logger, vdl_writer):
+
+    global_config = config['Global']
+
+    # build dataloader
+    valid_dataloader = build_dataloader(config, 'Eval', device, logger)
+
+    # build post process
+    post_process_class = build_post_process(config['PostProcess'],
+                                            global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, 'character'):
+        char_num = len(getattr(post_process_class, 'character'))
+        config['Architecture']["Head"]['out_channels'] = char_num
+    model = build_model(config['Architecture'])
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs before pruning: {flops}")
+
+    from paddleslim.dygraph import FPGMFilterPruner
+    model.train()
+    pruner = FPGMFilterPruner(model, [1, 3, 640, 640])
+
+    # build metric
+    eval_class = build_metric(config['Metric'])
+
+    def eval_fn():
+        metric = program.eval(model, valid_dataloader, post_process_class,
+                              eval_class)
+        logger.info(f"metric['hmean']: {metric['hmean']}")
+        return metric['hmean']
+
+    params_sensitive = pruner.sensitive(
+        eval_func=eval_fn,
+        sen_file="./sen.pickle",
+        skip_vars=[
+            "conv2d_57.w_0", "conv2d_transpose_2.w_0", "conv2d_transpose_3.w_0"
+        ])
+
+    logger.info(
+        "The sensitivity analysis results of model parameters saved in sen.pickle"
+    )
+    # calculate pruned params's ratio
+    params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+    for key in params_sensitive.keys():
+        logger.info(f"{key}, {params_sensitive[key]}")
+
+    plan = pruner.prune_vars(params_sensitive, [0])
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs after pruning: {flops}")
+
+    # load pretrain model
+    pre_best_model_dict = init_model(config, model, logger, None)
+    metric = program.eval(model, valid_dataloader, post_process_class,
+                          eval_class)
+    logger.info(f"metric['hmean']: {metric['hmean']}")
+
+    # start export model
+    from paddle.jit import to_static
+
+    infer_shape = [3, -1, -1]
+    if config['Architecture']['model_type'] == "rec":
+        infer_shape = [3, 32, -1]  # for rec model, H must be 32
+
+        if 'Transform' in config['Architecture'] and config['Architecture'][
+                'Transform'] is not None and config['Architecture'][
+                    'Transform']['name'] == 'TPS':
+            logger.info(
+                'When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training'
+            )
+            infer_shape[-1] = 100
+    model = to_static(
+        model,
+        input_spec=[
+            paddle.static.InputSpec(
+                shape=[None] + infer_shape, dtype='float32')
+        ])
+
+    save_path = '{}/inference'.format(config['Global']['save_inference_dir'])
+    paddle.jit.save(model, save_path)
+    logger.info('inference model is saved to {}'.format(save_path))
+
+
+if __name__ == '__main__':
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)

deploy/slim/prune/sensitivity_anal.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, '..', '..', '..'))
+sys.path.append(os.path.join(__dir__, '..', '..', '..', 'tools'))
+
+import paddle
+import paddle.distributed as dist
+from ppocr.data import build_dataloader
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import init_model
+import tools.program as program
+
+dist.get_world_size()
+
+
+def get_pruned_params(parameters):
+    params = []
+
+    for param in parameters:
+        if len(
+                param.shape
+        ) == 4 and 'depthwise' not in param.name and 'transpose' not in param.name and "conv2d_57" not in param.name and "conv2d_56" not in param.name:
+            params.append(param.name)
+    return params
+
+
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config['Global']['distributed']:
+        dist.init_parallel_env()
+
+    global_config = config['Global']
+
+    # build dataloader
+    train_dataloader = build_dataloader(config, 'Train', device, logger)
+    if config['Eval']:
+        valid_dataloader = build_dataloader(config, 'Eval', device, logger)
+    else:
+        valid_dataloader = None
+
+    # build post process
+    post_process_class = build_post_process(config['PostProcess'],
+                                            global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, 'character'):
+        char_num = len(getattr(post_process_class, 'character'))
+        config['Architecture']["Head"]['out_channels'] = char_num
+    model = build_model(config['Architecture'])
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs before pruning: {flops}")
+
+    from paddleslim.dygraph import FPGMFilterPruner
+    model.train()
+    pruner = FPGMFilterPruner(model, [1, 3, 640, 640])
+
+    # build loss
+    loss_class = build_loss(config['Loss'])
+
+    # build optim
+    optimizer, lr_scheduler = build_optimizer(
+        config['Optimizer'],
+        epochs=config['Global']['epoch_num'],
+        step_each_epoch=len(train_dataloader),
+        parameters=model.parameters())
+
+    # build metric
+    eval_class = build_metric(config['Metric'])
+    # load pretrain model
+    pre_best_model_dict = init_model(config, model, logger, optimizer)
+
+    logger.info('train dataloader has {} iters, valid dataloader has {} iters'.
+                format(len(train_dataloader), len(valid_dataloader)))
+    # build metric
+    eval_class = build_metric(config['Metric'])
+
+    logger.info('train dataloader has {} iters, valid dataloader has {} iters'.
+                format(len(train_dataloader), len(valid_dataloader)))
+
+    def eval_fn():
+        metric = program.eval(model, valid_dataloader, post_process_class,
+                              eval_class)
+        logger.info(f"metric['hmean']: {metric['hmean']}")
+        return metric['hmean']
+
+    params_sensitive = pruner.sensitive(
+        eval_func=eval_fn,
+        sen_file="./sen.pickle",
+        skip_vars=[
+            "conv2d_57.w_0", "conv2d_transpose_2.w_0", "conv2d_transpose_3.w_0"
+        ])
+
+    logger.info(
+        "The sensitivity analysis results of model parameters saved in sen.pickle"
+    )
+    # calculate pruned params's ratio
+    params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+    for key in params_sensitive.keys():
+        logger.info(f"{key}, {params_sensitive[key]}")
+
+    plan = pruner.prune_vars(params_sensitive, [0])
+    for param in model.parameters():
+        if ("weights" in param.name and "conv" in param.name) or (
+                "w_0" in param.name and "conv2d" in param.name):
+            logger.info(f"{param.name}: {param.shape}")
+
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs after pruning: {flops}")
+
+    # start train
+
+    program.train(config, train_dataloader, valid_dataloader, device, model,
+                  loss_class, optimizer, lr_scheduler, post_process_class,
+                  eval_class, pre_best_model_dict, logger, vdl_writer)
+
+
+if __name__ == '__main__':
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)