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

PPOCRLabel/PPOCRLabel.py

@@ -24,6 +24,7 @@ import sys
 from functools import partial
 from collections import defaultdict
 import json
+import cv2
 
 
 __dir__ = os.path.dirname(os.path.abspath(__file__))
@@ -1242,10 +1243,13 @@ class MainWindow(QMainWindow, WindowMixin):
         #     if unicodeFilePath in self.mImgList:
                 
         if unicodeFilePath and os.path.exists(unicodeFilePath):
-            self.imageData = read(unicodeFilePath, None)
             self.canvas.verified = False
 
-            image = QImage.fromData(self.imageData)
+            cvimg = cv2.imdecode(np.fromfile(unicodeFilePath, dtype=np.uint8), 1)
+            height, width, depth = cvimg.shape
+            cvimg = cv2.cvtColor(cvimg, cv2.COLOR_BGR2RGB)
+            image = QImage(cvimg.data, width, height, width * depth, QImage.Format_RGB888)
+
             if image.isNull():
                 self.errorMessage(u'Error opening file',
                                   u"<p>Make sure <i>%s</i> is a valid image file." % unicodeFilePath)

PPOCRLabel/libs/autoDialog.py

@@ -7,6 +7,8 @@ except ImportError:
     from PyQt4.QtCore import *
 
 import json
+import cv2
+import numpy as np
 
 from libs.utils import newIcon
 
@@ -34,11 +36,16 @@ class Worker(QThread):
                 if self.handle == 0:
                     self.listValue.emit(Imgpath)
                     if self.model == 'paddle':
-                        self.result_dic = self.ocr.ocr(Imgpath, cls=True, det=True)
+                        h, w, _ = cv2.imdecode(np.fromfile(Imgpath, dtype=np.uint8), 1).shape
+                        if h > 32 and w > 32:
+                            self.result_dic = self.ocr.ocr(Imgpath, cls=True, det=True)
+                        else:
+                            print('The size of', Imgpath, 'is too small to be recognised')
+                            self.result_dic = None
 
                     # 结果保存
                     if self.result_dic is None or len(self.result_dic) == 0:
-                        print('Can not recognise file  is :  ', Imgpath)
+                        print('Can not recognise file', Imgpath)
                         pass
                     else:
                         strs = ''

README_ch.md

@@ -8,7 +8,7 @@ PaddleOCR同时支持动态图与静态图两种编程范式
 - 静态图版本：develop分支
 
 **近期更新**
-- 2021.1.18 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数152个，每周一都会更新，欢迎大家持续关注。
+- 2021.1.25 [FAQ](./doc/doc_ch/FAQ.md)新增5个高频问题，总数157个，每周一都会更新，欢迎大家持续关注。
 - 2020.12.15 更新数据合成工具[Style-Text](./StyleText/README_ch.md)，可以批量合成大量与目标场景类似的图像，在多个场景验证，效果明显提升。
 - 2020.11.25 更新半自动标注工具[PPOCRLabel](./PPOCRLabel/README_ch.md)，辅助开发者高效完成标注任务，输出格式与PP-OCR训练任务完美衔接。
 - 2020.9.22 更新PP-OCR技术文章，https://arxiv.org/abs/2009.09941

deploy/docker/hubserving/cpu/Dockerfile

-# Version: 1.0.0
-FROM hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda10.0-cudnn7-dev
+# Version: 2.0.0
+FROM registry.baidubce.com/paddlepaddle/paddle:2.0.0rc1
 
 # PaddleOCR base on Python3.7
 RUN pip3.7 install --upgrade pip -i https://mirror.baidu.com/pypi/simple
 
-RUN python3.7 -m pip install paddlepaddle==2.0.0rc0 -i https://mirror.baidu.com/pypi/simple
-
 RUN pip3.7 install paddlehub --upgrade -i https://mirror.baidu.com/pypi/simple
 
 RUN git clone https://github.com/PaddlePaddle/PaddleOCR.git /PaddleOCR
@@ -15,15 +13,15 @@ WORKDIR /PaddleOCR
 RUN pip3.7 install -r requirements.txt -i https://mirror.baidu.com/pypi/simple
 
 RUN mkdir -p /PaddleOCR/inference/
-# Download orc detect model(light version). if you want to change normal version, you can change ch_ppocr_mobile_v1.1_det_infer to ch_ppocr_server_v1.1_det_infer, also remember change det_model_dir in deploy/hubserving/ocr_system/params.py）
+# Download orc detect model(light version). if you want to change normal version, you can change ch_ppocr_mobile_v2.0_det_infer to ch_ppocr_server_v2.0_det_infer, also remember change det_model_dir in deploy/hubserving/ocr_system/params.py）
 ADD {link} /PaddleOCR/inference/
 RUN tar xf /PaddleOCR/inference/{file} -C /PaddleOCR/inference/
 
-# Download direction classifier(light version). If you want to change normal version, you can change ch_ppocr_mobile_v1.1_cls_infer to ch_ppocr_mobile_v1.1_cls_infer, also remember change cls_model_dir in deploy/hubserving/ocr_system/params.py）
+# Download direction classifier(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_cls_infer to ch_ppocr_mobile_v2.0_cls_infer, also remember change cls_model_dir in deploy/hubserving/ocr_system/params.py）
 ADD {link} /PaddleOCR/inference/
 RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
 
-# Download orc recognition model(light version). If you want to change normal version, you can change ch_ppocr_mobile_v1.1_rec_infer to ch_ppocr_server_v1.1_rec_infer, also remember change rec_model_dir in deploy/hubserving/ocr_system/params.py）
+# Download orc recognition model(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_rec_infer to ch_ppocr_server_v2.0_rec_infer, also remember change rec_model_dir in deploy/hubserving/ocr_system/params.py）
 ADD {link} /PaddleOCR/inference/
 RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
 

deploy/docker/hubserving/gpu/Dockerfile

-# Version: 1.0.0
-FROM hub.baidubce.com/paddlepaddle/paddle:latest-gpu-cuda10.0-cudnn7-dev
+# Version: 2.0.0
+FROM egistry.baidubce.com/paddlepaddle/paddle:2.0.0rc1-gpu-cuda10.0-cudnn7
 
 # PaddleOCR base on Python3.7
 RUN pip3.7 install --upgrade pip -i https://mirror.baidu.com/pypi/simple
 
-RUN python3.7 -m pip install paddlepaddle-gpu==2.0.0rc0 -i https://mirror.baidu.com/pypi/simple
-
 RUN pip3.7 install paddlehub --upgrade -i https://mirror.baidu.com/pypi/simple
 
 RUN git clone https://github.com/PaddlePaddle/PaddleOCR.git /PaddleOCR
@@ -15,15 +13,15 @@ WORKDIR /PaddleOCR
 RUN pip3.7 install -r requirements.txt -i https://mirror.baidu.com/pypi/simple
 
 RUN mkdir -p /PaddleOCR/inference/
-# Download orc detect model(light version). if you want to change normal version, you can change ch_ppocr_mobile_v1.1_det_infer to ch_ppocr_server_v1.1_det_infer, also remember change det_model_dir in deploy/hubserving/ocr_system/params.py）
+# Download orc detect model(light version). if you want to change normal version, you can change ch_ppocr_mobile_v2.0_det_infer to ch_ppocr_server_v2.0_det_infer, also remember change det_model_dir in deploy/hubserving/ocr_system/params.py）
 ADD {link} /PaddleOCR/inference/
 RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
 
-# Download direction classifier(light version). If you want to change normal version, you can change ch_ppocr_mobile_v1.1_cls_infer to ch_ppocr_mobile_v1.1_cls_infer, also remember change cls_model_dir in deploy/hubserving/ocr_system/params.py）
+# Download direction classifier(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_cls_infer to ch_ppocr_mobile_v2.0_cls_infer, also remember change cls_model_dir in deploy/hubserving/ocr_system/params.py）
 ADD {link} /PaddleOCR/inference/
 RUN tar xf /PaddleOCR/inference/{file} -C /PaddleOCR/inference/
 
-# Download orc recognition model(light version). If you want to change normal version, you can change ch_ppocr_mobile_v1.1_rec_infer to ch_ppocr_server_v1.1_rec_infer, also remember change rec_model_dir in deploy/hubserving/ocr_system/params.py）
+# Download orc recognition model(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_rec_infer to ch_ppocr_server_v2.0_rec_infer, also remember change rec_model_dir in deploy/hubserving/ocr_system/params.py）
 ADD {link} /PaddleOCR/inference/
 RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
 

doc/doc_ch/FAQ.md

@@ -9,42 +9,43 @@
 
 ## PaddleOCR常见问题汇总(持续更新)
 
-* [近期更新（2021.1.18）](#近期更新)
+* [近期更新（2021.1.25）](#近期更新)
 * [【精选】OCR精选10个问题](#OCR精选10个问题)
 * [【理论篇】OCR通用32个问题](#OCR通用问题)
   * [基础知识7题](#基础知识)
   * [数据集7题](#数据集2)
   * [模型训练调优18题](#模型训练调优2)
-* [【实战篇】PaddleOCR实战110个问题](#PaddleOCR实战问题)
-  * [使用咨询36题](#使用咨询)
+* [【实战篇】PaddleOCR实战115个问题](#PaddleOCR实战问题)
+  * [使用咨询38题](#使用咨询)
   * [数据集17题](#数据集3)
   * [模型训练调优28题](#模型训练调优3)
-  * [预测部署29题](#预测部署3)
+  * [预测部署32题](#预测部署3)
 
 
 <a name="近期更新"></a>
-## 近期更新（2021.1.18）
+## 近期更新（2021.1.25）
 
+#### Q3.1.37: 小语种模型只有识别模型，没有检测模型吗？
 
-#### Q2.3.18: 在PP-OCR系统中，文本检测的骨干网络为什么没有使用SE模块？
+**A**：小语种（包括纯英文数字）的检测模型和中文的检测模型是共用的，在训练中文检测模型时加入了多语言数据。https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_en/models_list_en.md#1-text-detection-model。
 
-**A**：SE模块是MobileNetV3网络一个重要模块，目的是估计特征图每个特征通道重要性，给特征图每个特征分配权重，提高网络的表达能力。但是，对于文本检测，输入网络的分辨率比较大，一般是640\*640，利用SE模块估计特征图每个特征通道重要性比较困难，网络提升能力有限，但是该模块又比较耗时，因此在PP-OCR系统中，文本检测的骨干网络没有使用SE模块。实验也表明，当去掉SE模块，超轻量模型大小可以减小40%，文本检测效果基本不受影响。详细可以参考PP-OCR技术文章，https://arxiv.org/abs/2009.09941.
+#### Q3.1.38: module 'paddle.distributed' has no attribute ‘get_rank’。
 
-#### Q3.3.27: PaddleOCR关于文本识别模型的训练，支持的数据增强方式有哪些？
+**A**：Paddle版本问题，请安装2.0版本Paddle：pip install paddlepaddle==2.0.0rc1。
 
-**A**：文本识别支持的数据增强方式有随机小幅度裁剪、图像平衡、添加白噪声、颜色漂移、图像反色和Text Image Augmentation（TIA）变换等。可以参考[代码](../../ppocr/data/imaug/rec_img_aug.py)中的warp函数。
+#### Q3.4.30: PaddleOCR是否支持在华为鲲鹏920CPU上部署？
 
-#### Q3.3.28: 关于dygraph分支中，文本识别模型训练，要使用数据增强应该如何设置？
+**A**：目前Paddle的预测库是支持华为鲲鹏920CPU的，但是OCR还没在这些芯片上测试过，可以自己调试，有问题反馈给我们。
 
-**A**：可以参考[配置文件](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)在Train['dataset']['transforms']添加RecAug字段，使数据增强生效。可以通过添加对aug_prob设置，表示每种数据增强采用的概率。aug_prob默认是0.4.由于tia数据增强特殊性，默认不采用，可以通过添加use_tia设置，使tia数据增强生效。详细设置可以参考[ISSUE 1744](https://github.com/PaddlePaddle/PaddleOCR/issues/1744)。
+#### Q3.4.31: 采用Paddle-Lite进行端侧部署，出现问题，环境没问题。
 
-#### Q3.4.28: PP-OCR系统中，文本检测的结果有置信度吗？
+**A**：如果你的预测库是自己编译的，那么你的nb文件也要自己编译，用同一个lite版本。不能直接用下载的nb文件，因为版本不同。
 
-**A**：文本检测的结果有置信度，由于推理过程中没有使用，所以没有显示的返回到最终结果中。如果需要文本检测结果的置信度，可以在[文本检测DB的后处理代码](../../ppocr/postprocess/db_postprocess.py)的155行，添加scores信息。这样，在[检测预测代码](../../tools/infer/predict_det.py)的197行，就可以拿到文本检测的scores信息。
+#### Q3.4.32: PaddleOCR的模型支持onnx转换吗？
 
-#### Q3.4.29: DB文本检测，特征提取网络金字塔构建的部分代码在哪儿？
-
-**A**：特征提取网络金字塔构建的部分:[代码位置](../../ppocr/modeling/necks/db_fpn.py)。ppocr/modeling文件夹里面是组网相关的代码，其中architectures是文本检测或者文本识别整体流程代码；backbones是骨干网络相关代码；necks是类似与FPN的颈函数代码；heads是提取文本检测或者文本识别预测结果相关的头函数；transforms是类似于TPS特征预处理模块。更多的信息可以参考[代码组织结构](./tree.md)。
+**A**：我们目前已经通过Paddle2ONNX来支持各模型套件的转换，PaddleOCR基于PaddlePaddle 2.0的版本（dygraph分支）已经支持导出为ONNX，欢迎关注Paddle2ONNX，了解更多项目的进展：
+Paddle2ONNX项目：https://github.com/PaddlePaddle/Paddle2ONNX
+Paddle2ONNX支持转换的[模型列表](https://github.com/PaddlePaddle/Paddle2ONNX/blob/develop/docs/zh/model_zoo.md#%E5%9B%BE%E5%83%8Focr)
 
 <a name="OCR精选10个问题"></a>
 ## 【精选】OCR精选10个问题
@@ -474,9 +475,18 @@ StyleText的用途主要是：提取style_image中的字体、背景等style信
 例如识别身份证照片，可以先匹配"姓名"，"性别"等关键字，根据这些关键字的坐标去推测其他信息的位置，再与识别的结果匹配。
 
 #### Q3.1.36 如何识别竹简上的古文？
+
 **A**：对于字符都是普通的汉字字符的情况，只要标注足够的数据，finetune模型就可以了。如果数据量不足，您可以尝试StyleText工具。
 而如果使用的字符是特殊的古文字、甲骨文、象形文字等，那么首先需要构建一个古文字的字典，之后再进行训练。
 
+#### Q3.1.37: 小语种模型只有识别模型，没有检测模型吗？
+
+**A**：小语种（包括纯英文数字）的检测模型和中文的检测模型是共用的，在训练中文检测模型时加入了多语言数据。https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_en/models_list_en.md#1-text-detection-model。
+
+#### Q3.1.38: module 'paddle.distributed' has no attribute ‘get_rank’。
+
+**A**：Paddle版本问题，请安装2.0版本Paddle：pip install paddlepaddle==2.0.0rc1。
+
 <a name="数据集3"></a>
 ### 数据集
 
@@ -854,3 +864,17 @@ img = cv.imdecode(img_array, -1)
 #### Q3.4.29: DB文本检测，特征提取网络金字塔构建的部分代码在哪儿？
 
 **A**：特征提取网络金字塔构建的部分:[代码位置](../../ppocr/modeling/necks/db_fpn.py)。ppocr/modeling文件夹里面是组网相关的代码，其中architectures是文本检测或者文本识别整体流程代码；backbones是骨干网络相关代码；necks是类似与FPN的颈函数代码；heads是提取文本检测或者文本识别预测结果相关的头函数；transforms是类似于TPS特征预处理模块。更多的信息可以参考[代码组织结构](./tree.md)。
+
+#### Q3.4.30: PaddleOCR是否支持在华为鲲鹏920CPU上部署？
+
+**A**：目前Paddle的预测库是支持华为鲲鹏920CPU的，但是OCR还没在这些芯片上测试过，可以自己调试，有问题反馈给我们。
+
+#### Q3.4.31: 采用Paddle-Lite进行端侧部署，出现问题，环境没问题。
+
+**A**：如果你的预测库是自己编译的，那么你的nb文件也要自己编译，用同一个lite版本。不能直接用下载的nb文件，因为版本不同。
+
+#### Q3.4.32: PaddleOCR的模型支持onnx转换吗？
+
+**A**：我们目前已经通过Paddle2ONNX来支持各模型套件的转换，PaddleOCR基于PaddlePaddle 2.0的版本（dygraph分支）已经支持导出为ONNX，欢迎关注Paddle2ONNX，了解更多项目的进展：
+Paddle2ONNX项目：https://github.com/PaddlePaddle/Paddle2ONNX
+Paddle2ONNX支持转换的[模型列表](https://github.com/PaddlePaddle/Paddle2ONNX/blob/develop/docs/zh/model_zoo.md#%E5%9B%BE%E5%83%8Focr)
\ No newline at end of file

doc/doc_ch/algorithm_overview.md

@@ -14,11 +14,10 @@ PaddleOCR开源的文本检测算法列表：
 - [x]  SAST([paper](https://arxiv.org/abs/1908.05498))[4]
 
 在ICDAR2015文本检测公开数据集上，算法效果如下：
-
 |模型|骨干网络|precision|recall|Hmean|下载链接|
 | --- | --- | --- | --- | --- | --- |
-|EAST|ResNet50_vd|88.76%|81.36%|84.90%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
-|EAST|MobileNetV3|78.24%|79.15%|78.69%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
+|EAST|ResNet50_vd|85.80%|86.71%|86.25%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
+|EAST|MobileNetV3|79.42%|80.64%|80.03%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
 |DB|ResNet50_vd|86.41%|78.72%|82.38%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
 |DB|MobileNetV3|77.29%|73.08%|75.12%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
 |SAST|ResNet50_vd|91.39%|83.77%|87.42%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
@@ -40,17 +39,19 @@ PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训
 PaddleOCR基于动态图开源的文本识别算法列表：
 - [x]  CRNN([paper](https://arxiv.org/abs/1507.05717))[7]（ppocr推荐）
 - [x]  Rosetta([paper](https://arxiv.org/abs/1910.05085))[10]
-- [ ]  STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] coming soon
+- [x]  STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11]
 - [ ]  RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] coming soon
 - [ ]  SRN([paper](https://arxiv.org/abs/2003.12294))[5] coming soon
 
 参考[DTRB][3](https://arxiv.org/abs/1904.01906)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
 
 |模型|骨干网络|Avg Accuracy|模型存储命名|下载链接|
-|-|-|-|-|-|
+|---|---|---|---|---|
 |Rosetta|Resnet34_vd|80.9%|rec_r34_vd_none_none_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
 |Rosetta|MobileNetV3|78.05%|rec_mv3_none_none_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
 |CRNN|Resnet34_vd|82.76%|rec_r34_vd_none_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
 |CRNN|MobileNetV3|79.97%|rec_mv3_none_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
+|StarNet|Resnet34_vd|84.44%|rec_r34_vd_tps_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)|
+|StarNet|MobileNetV3|81.42%|rec_mv3_tps_bilstm_ctc|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)|
 
 PaddleOCR文本识别算法的训练和使用请参考文档教程中[模型训练/评估中的文本识别部分](./recognition.md)。

doc/doc_ch/inference.md

@@ -352,10 +352,10 @@ Predicts of ./doc/imgs_words/ch/word_4.jpg:['0', 0.9999982]
 
 ```
 # 使用方向分类器
-python3 tools/infer/predict_system.py --image_dir="./doc/imgs/2.jpg" --det_model_dir="./inference/det_db/" --cls_model_dir="./inference/cls/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=true
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./inference/det_db/" --cls_model_dir="./inference/cls/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=true
 
 # 不使用方向分类器
-python3 tools/infer/predict_system.py --image_dir="./doc/imgs/2.jpg" --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=false
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=false
 ```
 
 
@@ -364,7 +364,7 @@ python3 tools/infer/predict_system.py --image_dir="./doc/imgs/2.jpg" --det_model
 
 执行命令后，识别结果图像如下：
 
-![](../imgs_results/2.jpg)
+![](../imgs_results/system_res_00018069.jpg)
 
 <a name="其他模型推理"></a>
 ### 2. 其他模型推理
@@ -381,4 +381,4 @@ python3 tools/infer/predict_system.py --image_dir="./doc/imgs_en/img_10.jpg" --d
 
 执行命令后，识别结果图像如下：
 
-(coming soon)
+![](../imgs_results/img_10_east_starnet.jpg)

doc/doc_ch/models_list.md

-## OCR模型列表（V2.0，2020年12月12日更新）
+## OCR模型列表（V2.0，2021年1月20日更新）
 **说明** ：2.0版模型和[1.1版模型](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_ch/models_list.md)的主要区别在于动态图训练vs.静态图训练，模型性能上无明显差距。
 
 - [一、文本检测模型](#文本检测模型)
@@ -22,7 +22,7 @@ PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训
 
 |模型名称|模型简介|配置文件|推理模型大小|下载地址|
 | --- | --- | --- | --- | --- |
-|ch_ppocr_mobile_slim_v2.0_det|slim裁剪版超轻量模型，支持中英文、多语种文本检测|[ch_det_mv3_db_v2.0.yml](../../configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml)| |推理模型 (coming soon) / slim模型 (coming soon)|
+|ch_ppocr_mobile_slim_v2.0_det|slim裁剪版超轻量模型，支持中英文、多语种文本检测|[ch_det_mv3_db_v2.0.yml](../../configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml)| |推理模型 (coming soon) / 训练模型 (coming soon)|
 |ch_ppocr_mobile_v2.0_det|原始超轻量模型，支持中英文、多语种文本检测|[ch_det_mv3_db_v2.0.yml](../../configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml)|3M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar)|
 |ch_ppocr_server_v2.0_det|通用模型，支持中英文、多语种文本检测，比超轻量模型更大，但效果更好|[ch_det_res18_db_v2.0.yml](../../configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml)|47M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar)|
 
@@ -35,7 +35,7 @@ PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训
 
 |模型名称|模型简介|配置文件|推理模型大小|下载地址|
 | --- | --- | --- | --- | --- |
-|ch_ppocr_mobile_slim_v2.0_rec|slim裁剪量化版超轻量模型，支持中英文、数字识别|[rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)| |推理模型 (coming soon) / slim模型 (coming soon) |
+|ch_ppocr_mobile_slim_v2.0_rec|slim裁剪量化版超轻量模型，支持中英文、数字识别|[rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)| |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_train.tar) |
 |ch_ppocr_mobile_v2.0_rec|原始超轻量模型，支持中英文、数字识别|[rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)|3.71M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) |
 |ch_ppocr_server_v2.0_rec|通用模型，支持中英文、数字识别|[rec_chinese_common_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml)|94.8M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) |
 
@@ -46,7 +46,7 @@ PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训
 
 |模型名称|模型简介|配置文件|推理模型大小|下载地址|
 | --- | --- | --- | --- | --- |
-|en_number_mobile_slim_v2.0_rec|slim裁剪量化版超轻量模型，支持英文、数字识别|[rec_en_number_lite_train.yml](../../configs/rec/multi_language/rec_en_number_lite_train.yml)| | 推理模型 (coming soon) / slim模型 (coming soon) |
+|en_number_mobile_slim_v2.0_rec|slim裁剪量化版超轻量模型，支持英文、数字识别|[rec_en_number_lite_train.yml](../../configs/rec/multi_language/rec_en_number_lite_train.yml)| | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_train.tar) |
 |en_number_mobile_v2.0_rec|原始超轻量模型，支持英文、数字识别|[rec_en_number_lite_train.yml](../../configs/rec/multi_language/rec_en_number_lite_train.yml)|2.56M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_train.tar) |
 
 <a name="多语言识别模型"></a>
@@ -55,7 +55,7 @@ PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训
 **说明：** 新增的多语言模型的配置文件通过代码方式生成，您可以通过`--help`参数查看当前PaddleOCR支持生成哪些多语言的配置文件：
 ```bash
 # 该代码需要在指定目录运行
-cd PaddleOCR/configs/rec/multi_language/
+cd {your/path/}PaddleOCR/configs/rec/multi_language/
 python3 generate_multi_language_configs.py --help
 ```
 下面以生成意大利语配置文件为例：
@@ -64,7 +64,7 @@ python3 generate_multi_language_configs.py --help
 如果您仅仅想用配置文件测试PaddleOCR提供的多语言模型可以通过下面命令生成默认的配置文件，使用PaddleOCR提供的小语种字典进行预测。
 ```bash
 # 该代码需要在指定目录运行
-cd PaddleOCR/configs/rec/multi_language/
+cd {your/path/}PaddleOCR/configs/rec/multi_language/
 # 通过-l或者--language参数设置需要生成的语种的配置文件，该命令会将默认参数写入配置文件
 python3 generate_multi_language_configs.py -l it
 ```
@@ -77,6 +77,8 @@ python3 generate_multi_language_configs.py -l it
 
 使用以下命令生成配置文件：
 ```bash
+# 该代码需要在指定目录运行
+cd {your/path/}PaddleOCR/configs/rec/multi_language/
 # -l或者--language字段是必须的
 # --train修改训练集，--val修改验证集，--data_dir修改数据集目录，-o修改对应默认参数
 # --dict命令改变字典路径，示例使用默认字典路径则该参数可不填
@@ -112,7 +114,7 @@ python3 generate_multi_language_configs.py -l it \
 | uk_mobile_v2.0_rec |乌克兰文识别|rec_uk_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_train.tar) |
 | be_mobile_v2.0_rec |白俄罗斯文识别|rec_be_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_train.tar) |
 | te_mobile_v2.0_rec |泰卢固文识别|rec_te_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_train.tar) |
-| ka_mobile_v2.0_rec |卡纳达文识别|[rec_ka_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_train.tar) |
+| ka_mobile_v2.0_rec |卡纳达文识别|rec_ka_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_train.tar) |
 | ta_mobile_v2.0_rec |泰米尔文识别|rec_ta_lite_train.yml|2.63M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_train.tar) |
 
 
@@ -121,5 +123,5 @@ python3 generate_multi_language_configs.py -l it \
 
 |模型名称|模型简介|配置文件|推理模型大小|下载地址|
 | --- | --- | --- | --- | --- |
-|ch_ppocr_mobile_slim_v2.0_cls|slim量化版模型|[cls_mv3.yml](../../configs/cls/cls_mv3.yml)| |推理模型 (coming soon) / 训练模型 / slim模型 |
+|ch_ppocr_mobile_slim_v2.0_cls|slim量化版模型|[cls_mv3.yml](../../configs/cls/cls_mv3.yml)| |[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) |
 |ch_ppocr_mobile_v2.0_cls|原始模型|[cls_mv3.yml](../../configs/cls/cls_mv3.yml)|1.38M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |

doc/doc_en/algorithm_overview_en.md

@@ -19,8 +19,8 @@ On the ICDAR2015 dataset, the text detection result is as follows:
 
 |Model|Backbone|precision|recall|Hmean|Download link|
 | --- | --- | --- | --- | --- | --- |
-|EAST|ResNet50_vd|88.76%|81.36%|84.90%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
-|EAST|MobileNetV3|78.24%|79.15%|78.69%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
+|EAST|ResNet50_vd|85.80%|86.71%|86.25%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
+|EAST|MobileNetV3|79.42%|80.64%|80.03%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
 |DB|ResNet50_vd|86.41%|78.72%|82.38%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
 |DB|MobileNetV3|77.29%|73.08%|75.12%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
 |SAST|ResNet50_vd|91.39%|83.77%|87.42%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
@@ -41,17 +41,19 @@ For the training guide and use of PaddleOCR text detection algorithms, please re
 PaddleOCR open-source text recognition algorithms list:
 - [x]  CRNN([paper](https://arxiv.org/abs/1507.05717))[7]
 - [x]  Rosetta([paper](https://arxiv.org/abs/1910.05085))[10]
-- [ ]  STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11] coming soon
+- [x]  STAR-Net([paper](http://www.bmva.org/bmvc/2016/papers/paper043/index.html))[11]
 - [ ]  RARE([paper](https://arxiv.org/abs/1603.03915v1))[12] coming soon
 - [ ]  SRN([paper](https://arxiv.org/abs/2003.12294))[5] coming soon
 
 Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow:
 
 |Model|Backbone|Avg Accuracy|Module combination|Download link|
-|-|-|-|-|-|
+|---|---|---|---|---|
 |Rosetta|Resnet34_vd|80.9%|rec_r34_vd_none_none_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
 |Rosetta|MobileNetV3|78.05%|rec_mv3_none_none_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
 |CRNN|Resnet34_vd|82.76%|rec_r34_vd_none_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
 |CRNN|MobileNetV3|79.97%|rec_mv3_none_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
+|StarNet|Resnet34_vd|84.44%|rec_r34_vd_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)|
+|StarNet|MobileNetV3|81.42%|rec_mv3_tps_bilstm_ctc|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)|
 
 Please refer to the document for training guide and use of PaddleOCR text recognition algorithms [Text recognition model training/evaluation/prediction](./recognition_en.md)

doc/doc_en/inference_en.md

@@ -366,15 +366,15 @@ When performing prediction, you need to specify the path of a single image or a
 
 ```
 # use direction classifier
-python3 tools/infer/predict_system.py --image_dir="./doc/imgs/2.jpg" --det_model_dir="./inference/det_db/" --cls_model_dir="./inference/cls/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=true
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./inference/det_db/" --cls_model_dir="./inference/cls/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=true
 
 # not use use direction classifier
-python3 tools/infer/predict_system.py --image_dir="./doc/imgs/2.jpg" --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/"
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/"
 ```
 
 After executing the command, the recognition result image is as follows:
 
-![](../imgs_results/2.jpg)
+![](../imgs_results/system_res_00018069.jpg)
 
 <a name="OTHER_MODELS"></a>
 ### 2. OTHER MODELS
@@ -391,4 +391,4 @@ python3 tools/infer/predict_system.py --image_dir="./doc/imgs_en/img_10.jpg" --d
 
 After executing the command, the recognition result image is as follows:
 
-(coming soon)
+![](../imgs_results/img_10_east_starnet.jpg)

doc/doc_en/models_list_en.md

@@ -33,7 +33,7 @@ The downloadable models provided by PaddleOCR include `inference model`, `traine
 
 |model name|description|config|model size|download|
 | --- | --- | --- | --- | --- |
-|ch_ppocr_mobile_slim_v2.0_rec|Slim pruned and quantized lightweight model, supporting Chinese, English and number recognition|[rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)| |inference model (coming soon) / slim model (coming soon) |
+|ch_ppocr_mobile_slim_v2.0_rec|Slim pruned and quantized lightweight model, supporting Chinese, English and number recognition|[rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)| | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_train.tar) |
 |ch_ppocr_mobile_v2.0_rec|Original lightweight model, supporting Chinese, English and number recognition|[rec_chinese_lite_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)|3.71M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) |
 |ch_ppocr_server_v2.0_rec|General model, supporting Chinese, English and number recognition|[rec_chinese_common_train_v2.0.yml](../../configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml)|94.8M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) |
 
@@ -45,7 +45,7 @@ The downloadable models provided by PaddleOCR include `inference model`, `traine
 
 |model name|description|config|model size|download|
 | --- | --- | --- | --- | --- |
-|en_number_mobile_slim_v2.0_rec|Slim pruned and quantized lightweight model, supporting English and number recognition|[rec_en_number_lite_train.yml](../../configs/rec/multi_language/rec_en_number_lite_train.yml)| |inference model (coming soon ) / slim model (coming soon) |
+|en_number_mobile_slim_v2.0_rec|Slim pruned and quantized lightweight model, supporting English and number recognition|[rec_en_number_lite_train.yml](../../configs/rec/multi_language/rec_en_number_lite_train.yml)| | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_train.tar) |
 |en_number_mobile_v2.0_rec|Original lightweight model, supporting English and number recognition|[rec_en_number_lite_train.yml](../../configs/rec/multi_language/rec_en_number_lite_train.yml)|2.56M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_train.tar) |
 
 <a name="Multilingual"></a>
@@ -54,6 +54,8 @@ The downloadable models provided by PaddleOCR include `inference model`, `traine
 **Note：** The configuration file of the new multi language model is generated by code. You can use the `--help` parameter to check which multi language are supported by current PaddleOCR.
 
 ```bash
+# The code needs to run in the specified directory
+cd {your/path/}PaddleOCR/configs/rec/multi_language/
 python3 generate_multi_language_configs.py --help
 ```
 
@@ -62,7 +64,7 @@ Take the Italian configuration file as an example：
 you can generate the default configuration file through the following command, and use the default language dictionary provided by paddleocr for prediction.
 ```bash
 # The code needs to run in the specified directory
-cd PaddleOCR/configs/rec/multi_language/
+cd {your/path/}PaddleOCR/configs/rec/multi_language/
 # Set the required language configuration file through -l or --language parameter
 # This command will write the default parameter to the configuration file.
 python3 generate_multi_language_configs.py -l it
@@ -74,6 +76,8 @@ If you want to train your own model, you can prepare the training set file, veri
 - Use the default dictionary provided by paddleocr:{your/path/}PaddleOCR/ppocr/utils/dict/it_dict.txt
 - Training data path:{your/path/}PaddleOCR/train_data
 ```bash
+# The code needs to run in the specified directory
+cd {your/path/}PaddleOCR/configs/rec/multi_language/
 # The -l or --language parameter is required
 # --train modify train_list path
 # --val modify eval_list path 
@@ -92,27 +96,27 @@ python3 generate_multi_language_configs.py -l it \
 | german_mobile_v2.0_rec |Lightweight model for French recognition|[rec_german_lite_train.yml](../../configs/rec/multi_language/rec_german_lite_train.yml)|2.65M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/german_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/german_mobile_v2.0_rec_train.tar) |
 | korean_mobile_v2.0_rec |Lightweight model for Korean recognition|[rec_korean_lite_train.yml](../../configs/rec/multi_language/rec_korean_lite_train.yml)|3.9M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/korean_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/korean_mobile_v2.0_rec_train.tar) |
 | japan_mobile_v2.0_rec |Lightweight model for Japanese recognition|[rec_japan_lite_train.yml](../../configs/rec/multi_language/rec_japan_lite_train.yml)|4.23M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/japan_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/japan_mobile_v2.0_rec_train.tar) |
-| it_mobile_v2.0_rec |Lightweight model for Italian recognition|rec_it_lite_train.yml|2.53M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/it_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/it_mobile_v2.0_rec_train.tar) |
-| xi_mobile_v2.0_rec |Lightweight model for Spanish recognition|rec_xi_lite_train.yml|2.53M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/xi_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/xi_mobile_v2.0_rec_train.tar) |
-| pu_mobile_v2.0_rec |Lightweight model for Portuguese recognition|rec_pu_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/pu_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/pu_mobile_v2.0_rec_train.tar) |
-| ru_mobile_v2.0_rec |Lightweight model for Russia recognition|rec_ru_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ru_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ru_mobile_v2.0_rec_train.tar) |
-| ar_mobile_v2.0_rec |Lightweight model for Arabic recognition|rec_ar_lite_train.yml|2.53M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ar_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ar_mobile_v2.0_rec_train.tar) |
-| hi_mobile_v2.0_rec |Lightweight model for Hindi recognition|rec_hi_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/hi_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/hi_mobile_v2.0_rec_train.tar) |
-| chinese_cht_mobile_v2.0_rec |Lightweight model for chinese traditional recognition|rec_chinese_cht_lite_train.yml|5.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/chinese_cht_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/chinese_cht_mobile_v2.0_rec_train.tar) |
-| ug_mobile_v2.0_rec |Lightweight model for Uyghur recognition|rec_ug_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ug_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ug_mobile_v2.0_rec_train.tar) |
-| fa_mobile_v2.0_rec |Lightweight model for Persian recognition|rec_fa_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/fa_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/fa_mobile_v2.0_rec_train.tar) |
-| ur_mobile_v2.0_rec |Lightweight model for Urdu recognition|rec_ur_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ur_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ur_mobile_v2.0_rec_train.tar) |
-| rs_mobile_v2.0_rec |Lightweight model for Serbian(latin) recognition|rec_rs_lite_train.yml|2.53M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rs_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rs_mobile_v2.0_rec_train.tar) |
-| oc_mobile_v2.0_rec |Lightweight model for Occitan recognition|rec_oc_lite_train.yml|2.53M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/oc_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/oc_mobile_v2.0_rec_train.tar) |
-| mr_mobile_v2.0_rec |Lightweight model for Marathi recognition|rec_mr_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/mr_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/mr_mobile_v2.0_rec_train.tar) |
-| ne_mobile_v2.0_rec |Lightweight model for Nepali recognition|rec_ne_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ne_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ne_mobile_v2.0_rec_train.tar) |
-| rsc_mobile_v2.0_rec |Lightweight model for Serbian(cyrillic) recognition|rec_rsc_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rsc_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rsc_mobile_v2.0_rec_train.tar) |
-| bg_mobile_v2.0_rec |Lightweight model for Bulgarian recognition|rec_bg_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/bg_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/bg_mobile_v2.0_rec_train.tar) |
-| uk_mobile_v2.0_rec |Lightweight model for Ukranian recognition|rec_uk_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_train.tar) |
-| be_mobile_v2.0_rec |Lightweight model for Belarusian recognition|rec_be_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_train.tar) |
-| te_mobile_v2.0_rec |Lightweight model for Telugu recognition|rec_te_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_train.tar) |
-| ka_mobile_v2.0_rec |Lightweight model for Kannada recognition|[rec_ka_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_train.tar) |
-| ta_mobile_v2.0_rec |Lightweight model for Tamil recognition|rec_ta_lite_train.yml|2.63M|[inference mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_infer.tar) / [trained mode](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_train.tar) |
+| it_mobile_v2.0_rec |Lightweight model for Italian recognition|rec_it_lite_train.yml|2.53M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/it_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/it_mobile_v2.0_rec_train.tar) |
+| xi_mobile_v2.0_rec |Lightweight model for Spanish recognition|rec_xi_lite_train.yml|2.53M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/xi_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/xi_mobile_v2.0_rec_train.tar) |
+| pu_mobile_v2.0_rec |Lightweight model for Portuguese recognition|rec_pu_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/pu_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/pu_mobile_v2.0_rec_train.tar) |
+| ru_mobile_v2.0_rec |Lightweight model for Russia recognition|rec_ru_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ru_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ru_mobile_v2.0_rec_train.tar) |
+| ar_mobile_v2.0_rec |Lightweight model for Arabic recognition|rec_ar_lite_train.yml|2.53M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ar_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ar_mobile_v2.0_rec_train.tar) |
+| hi_mobile_v2.0_rec |Lightweight model for Hindi recognition|rec_hi_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/hi_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/hi_mobile_v2.0_rec_train.tar) |
+| chinese_cht_mobile_v2.0_rec |Lightweight model for chinese traditional recognition|rec_chinese_cht_lite_train.yml|5.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/chinese_cht_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/chinese_cht_mobile_v2.0_rec_train.tar) |
+| ug_mobile_v2.0_rec |Lightweight model for Uyghur recognition|rec_ug_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ug_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ug_mobile_v2.0_rec_train.tar) |
+| fa_mobile_v2.0_rec |Lightweight model for Persian recognition|rec_fa_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/fa_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/fa_mobile_v2.0_rec_train.tar) |
+| ur_mobile_v2.0_rec |Lightweight model for Urdu recognition|rec_ur_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ur_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ur_mobile_v2.0_rec_train.tar) |
+| rs_mobile_v2.0_rec |Lightweight model for Serbian(latin) recognition|rec_rs_lite_train.yml|2.53M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rs_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rs_mobile_v2.0_rec_train.tar) |
+| oc_mobile_v2.0_rec |Lightweight model for Occitan recognition|rec_oc_lite_train.yml|2.53M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/oc_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/oc_mobile_v2.0_rec_train.tar) |
+| mr_mobile_v2.0_rec |Lightweight model for Marathi recognition|rec_mr_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/mr_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/mr_mobile_v2.0_rec_train.tar) |
+| ne_mobile_v2.0_rec |Lightweight model for Nepali recognition|rec_ne_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ne_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ne_mobile_v2.0_rec_train.tar) |
+| rsc_mobile_v2.0_rec |Lightweight model for Serbian(cyrillic) recognition|rec_rsc_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rsc_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/rsc_mobile_v2.0_rec_train.tar) |
+| bg_mobile_v2.0_rec |Lightweight model for Bulgarian recognition|rec_bg_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/bg_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/bg_mobile_v2.0_rec_train.tar) |
+| uk_mobile_v2.0_rec |Lightweight model for Ukranian recognition|rec_uk_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/uk_mobile_v2.0_rec_train.tar) |
+| be_mobile_v2.0_rec |Lightweight model for Belarusian recognition|rec_be_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/be_mobile_v2.0_rec_train.tar) |
+| te_mobile_v2.0_rec |Lightweight model for Telugu recognition|rec_te_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/te_mobile_v2.0_rec_train.tar) |
+| ka_mobile_v2.0_rec |Lightweight model for Kannada recognition|rec_ka_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ka_mobile_v2.0_rec_train.tar) |
+| ta_mobile_v2.0_rec |Lightweight model for Tamil recognition|rec_ta_lite_train.yml|2.63M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/ta_mobile_v2.0_rec_train.tar) |
 
 
 <a name="Angle"></a>
@@ -120,6 +124,5 @@ python3 generate_multi_language_configs.py -l it \
 
 |model name|description|config|model size|download|
 | --- | --- | --- | --- | --- |
-|ch_ppocr_mobile_slim_v2.0_cls|Slim quantized model|[cls_mv3.yml](../../configs/cls/cls_mv3.yml)| |inference model (coming soon) / trained model / slim model|
+|ch_ppocr_mobile_slim_v2.0_cls|Slim quantized model|[cls_mv3.yml](../../configs/cls/cls_mv3.yml)| | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_slim_train.tar) |
 |ch_ppocr_mobile_v2.0_cls|Original model|[cls_mv3.yml](../../configs/cls/cls_mv3.yml)|1.38M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |
-

paddleocr.py

@@ -290,7 +290,9 @@ class PaddleOCR(predict_system.TextSystem):
             image_file = img
             img, flag = check_and_read_gif(image_file)
             if not flag:
-                img = cv2.imread(image_file)
+                with open(image_file, 'rb') as f:
+                    np_arr = np.frombuffer(f.read(), dtype=np.uint8)
+                    img = cv2.imdecode(np_arr, cv2.IMREAD_COLOR)
             if img is None:
                 logger.error("error in loading image:{}".format(image_file))
                 return None

ppocr/data/__init__.py

@@ -51,7 +51,7 @@ signal.signal(signal.SIGINT, term_mp)
 signal.signal(signal.SIGTERM, term_mp)
 
 
-def build_dataloader(config, mode, device, logger):
+def build_dataloader(config, mode, device, logger, seed=None):
     config = copy.deepcopy(config)
 
     support_dict = ['SimpleDataSet', 'LMDBDateSet']
@@ -61,7 +61,7 @@ def build_dataloader(config, mode, device, logger):
     assert mode in ['Train', 'Eval', 'Test'
                     ], "Mode should be Train, Eval or Test."
 
-    dataset = eval(module_name)(config, mode, logger)
+    dataset = eval(module_name)(config, mode, logger, seed)
     loader_config = config[mode]['loader']
     batch_size = loader_config['batch_size_per_card']
     drop_last = loader_config['drop_last']

ppocr/data/imaug/sast_process.py

@@ -24,11 +24,11 @@ __all__ = ['SASTProcessTrain']
 
 class SASTProcessTrain(object):
     def __init__(self,
-                 image_shape = [512, 512],
-                 min_crop_size = 24,
-                 min_crop_side_ratio = 0.3,
-                 min_text_size = 10,
-                 max_text_size = 512,
+                 image_shape=[512, 512],
+                 min_crop_size=24,
+                 min_crop_side_ratio=0.3,
+                 min_text_size=10,
+                 max_text_size=512,
                  **kwargs):
         self.input_size = image_shape[1]
         self.min_crop_size = min_crop_size
@@ -42,12 +42,10 @@ class SASTProcessTrain(object):
         :param poly:
         :return:
         """
-        edge = [
-            (poly[1][0] - poly[0][0]) * (poly[1][1] + poly[0][1]),
-            (poly[2][0] - poly[1][0]) * (poly[2][1] + poly[1][1]),
-            (poly[3][0] - poly[2][0]) * (poly[3][1] + poly[2][1]),
-            (poly[0][0] - poly[3][0]) * (poly[0][1] + poly[3][1])
-        ]
+        edge = [(poly[1][0] - poly[0][0]) * (poly[1][1] + poly[0][1]),
+                (poly[2][0] - poly[1][0]) * (poly[2][1] + poly[1][1]),
+                (poly[3][0] - poly[2][0]) * (poly[3][1] + poly[2][1]),
+                (poly[0][0] - poly[3][0]) * (poly[0][1] + poly[3][1])]
         return np.sum(edge) / 2.
 
     def gen_quad_from_poly(self, poly):
@@ -57,7 +55,8 @@ class SASTProcessTrain(object):
         point_num = poly.shape[0]
         min_area_quad = np.zeros((4, 2), dtype=np.float32)
         if True:
-            rect = cv2.minAreaRect(poly.astype(np.int32))  # (center (x,y), (width, height), angle of rotation)
+            rect = cv2.minAreaRect(poly.astype(
+                np.int32))  # (center (x,y), (width, height), angle of rotation)
             center_point = rect[0]
             box = np.array(cv2.boxPoints(rect))
 
@@ -102,23 +101,33 @@ class SASTProcessTrain(object):
             if p_area > 0:
                 if tag == False:
                     print('poly in wrong direction')
-                    tag = True # reversed cases should be ignore
-                poly = poly[(0, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1), :]
+                    tag = True  # reversed cases should be ignore
+                poly = poly[(0, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2,
+                             1), :]
                 quad = quad[(0, 3, 2, 1), :]
 
-            len_w = np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[3] - quad[2])
-            len_h = np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[1] - quad[2])
+            len_w = np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[3] -
+                                                                       quad[2])
+            len_h = np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[1] -
+                                                                       quad[2])
             hv_tag = 1
-        
-            if len_w * 2.0 <  len_h:
+
+            if len_w * 2.0 < len_h:
                 hv_tag = 0
 
             validated_polys.append(poly)
             validated_tags.append(tag)
             hv_tags.append(hv_tag)
-        return np.array(validated_polys), np.array(validated_tags), np.array(hv_tags)
+        return np.array(validated_polys), np.array(validated_tags), np.array(
+            hv_tags)
 
-    def crop_area(self, im, polys, tags, hv_tags, crop_background=False, max_tries=25):
+    def crop_area(self,
+                  im,
+                  polys,
+                  tags,
+                  hv_tags,
+                  crop_background=False,
+                  max_tries=25):
         """
         make random crop from the input image
         :param im:
@@ -137,10 +146,10 @@ class SASTProcessTrain(object):
             poly = np.round(poly, decimals=0).astype(np.int32)
             minx = np.min(poly[:, 0])
             maxx = np.max(poly[:, 0])
-            w_array[minx + pad_w: maxx + pad_w] = 1
+            w_array[minx + pad_w:maxx + pad_w] = 1
             miny = np.min(poly[:, 1])
             maxy = np.max(poly[:, 1])
-            h_array[miny + pad_h: maxy + pad_h] = 1
+            h_array[miny + pad_h:maxy + pad_h] = 1
         # ensure the cropped area not across a text
         h_axis = np.where(h_array == 0)[0]
         w_axis = np.where(w_array == 0)[0]
@@ -166,17 +175,18 @@ class SASTProcessTrain(object):
             if polys.shape[0] != 0:
                 poly_axis_in_area = (polys[:, :, 0] >= xmin) & (polys[:, :, 0] <= xmax) \
                                     & (polys[:, :, 1] >= ymin) & (polys[:, :, 1] <= ymax)
-                selected_polys = np.where(np.sum(poly_axis_in_area, axis=1) == 4)[0]
+                selected_polys = np.where(
+                    np.sum(poly_axis_in_area, axis=1) == 4)[0]
             else:
                 selected_polys = []
             if len(selected_polys) == 0:
                 # no text in this area
                 if crop_background:
                     return im[ymin : ymax + 1, xmin : xmax + 1, :], \
-                        polys[selected_polys], tags[selected_polys], hv_tags[selected_polys], txts
+                        polys[selected_polys], tags[selected_polys], hv_tags[selected_polys]
                 else:
                     continue
-            im = im[ymin: ymax + 1, xmin: xmax + 1, :]
+            im = im[ymin:ymax + 1, xmin:xmax + 1, :]
             polys = polys[selected_polys]
             tags = tags[selected_polys]
             hv_tags = hv_tags[selected_polys]
@@ -192,18 +202,28 @@ class SASTProcessTrain(object):
         width_list = []
         height_list = []
         for quad in poly_quads:
-            quad_w = (np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[2] - quad[3])) / 2.0
-            quad_h = (np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[2] - quad[1])) / 2.0
+            quad_w = (np.linalg.norm(quad[0] - quad[1]) +
+                      np.linalg.norm(quad[2] - quad[3])) / 2.0
+            quad_h = (np.linalg.norm(quad[0] - quad[3]) +
+                      np.linalg.norm(quad[2] - quad[1])) / 2.0
             width_list.append(quad_w)
             height_list.append(quad_h)
-        norm_width = max(sum(width_list) / (len(width_list) +  1e-6), 1.0)
+        norm_width = max(sum(width_list) / (len(width_list) + 1e-6), 1.0)
         average_height = max(sum(height_list) / (len(height_list) + 1e-6), 1.0)
 
         for quad in poly_quads:
-            direct_vector_full = ((quad[1] + quad[2]) - (quad[0] + quad[3])) / 2.0
-            direct_vector = direct_vector_full / (np.linalg.norm(direct_vector_full) + 1e-6) * norm_width
-            direction_label = tuple(map(float, [direct_vector[0], direct_vector[1], 1.0 / (average_height + 1e-6)]))
-            cv2.fillPoly(direction_map, quad.round().astype(np.int32)[np.newaxis, :, :], direction_label)
+            direct_vector_full = (
+                (quad[1] + quad[2]) - (quad[0] + quad[3])) / 2.0
+            direct_vector = direct_vector_full / (
+                np.linalg.norm(direct_vector_full) + 1e-6) * norm_width
+            direction_label = tuple(
+                map(float, [
+                    direct_vector[0], direct_vector[1], 1.0 / (average_height +
+                                                               1e-6)
+                ]))
+            cv2.fillPoly(direction_map,
+                         quad.round().astype(np.int32)[np.newaxis, :, :],
+                         direction_label)
         return direction_map
 
     def calculate_average_height(self, poly_quads):
@@ -211,13 +231,19 @@ class SASTProcessTrain(object):
         """
         height_list = []
         for quad in poly_quads:
-            quad_h = (np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[2] - quad[1])) / 2.0
+            quad_h = (np.linalg.norm(quad[0] - quad[3]) +
+                      np.linalg.norm(quad[2] - quad[1])) / 2.0
             height_list.append(quad_h)
         average_height = max(sum(height_list) / len(height_list), 1.0)
         return average_height
 
-    def generate_tcl_label(self, hw, polys, tags, ds_ratio,
-                            tcl_ratio=0.3, shrink_ratio_of_width=0.15):
+    def generate_tcl_label(self,
+                           hw,
+                           polys,
+                           tags,
+                           ds_ratio,
+                           tcl_ratio=0.3,
+                           shrink_ratio_of_width=0.15):
         """
         Generate polygon.
         """
@@ -225,21 +251,30 @@ class SASTProcessTrain(object):
         h, w = int(h * ds_ratio), int(w * ds_ratio)
         polys = polys * ds_ratio
 
-        score_map = np.zeros((h, w,), dtype=np.float32)
+        score_map = np.zeros(
+            (
+                h,
+                w, ), dtype=np.float32)
         tbo_map = np.zeros((h, w, 5), dtype=np.float32)
-        training_mask = np.ones((h, w,), dtype=np.float32)
-        direction_map = np.ones((h, w, 3)) * np.array([0, 0, 1]).reshape([1, 1, 3]).astype(np.float32)
+        training_mask = np.ones(
+            (
+                h,
+                w, ), dtype=np.float32)
+        direction_map = np.ones((h, w, 3)) * np.array([0, 0, 1]).reshape(
+            [1, 1, 3]).astype(np.float32)
 
         for poly_idx, poly_tag in enumerate(zip(polys, tags)):
-            poly = poly_tag[0] 
+            poly = poly_tag[0]
             tag = poly_tag[1]
 
             # generate min_area_quad
             min_area_quad, center_point = self.gen_min_area_quad_from_poly(poly)
-            min_area_quad_h = 0.5 * (np.linalg.norm(min_area_quad[0] - min_area_quad[3]) +
-                                    np.linalg.norm(min_area_quad[1] - min_area_quad[2]))
-            min_area_quad_w = 0.5 * (np.linalg.norm(min_area_quad[0] - min_area_quad[1]) +
-                                    np.linalg.norm(min_area_quad[2] - min_area_quad[3]))
+            min_area_quad_h = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[3]) +
+                np.linalg.norm(min_area_quad[1] - min_area_quad[2]))
+            min_area_quad_w = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[1]) +
+                np.linalg.norm(min_area_quad[2] - min_area_quad[3]))
 
             if min(min_area_quad_h, min_area_quad_w) < self.min_text_size * ds_ratio \
                 or min(min_area_quad_h, min_area_quad_w) > self.max_text_size * ds_ratio:
@@ -247,25 +282,37 @@ class SASTProcessTrain(object):
 
             if tag:
                 # continue
-                cv2.fillPoly(training_mask, poly.astype(np.int32)[np.newaxis, :, :], 0.15)
+                cv2.fillPoly(training_mask,
+                             poly.astype(np.int32)[np.newaxis, :, :], 0.15)
             else:
                 tcl_poly = self.poly2tcl(poly, tcl_ratio)
                 tcl_quads = self.poly2quads(tcl_poly)
                 poly_quads = self.poly2quads(poly)
                 # stcl map
-                stcl_quads, quad_index = self.shrink_poly_along_width(tcl_quads, shrink_ratio_of_width=shrink_ratio_of_width,
-                                                                expand_height_ratio=1.0 / tcl_ratio)
+                stcl_quads, quad_index = self.shrink_poly_along_width(
+                    tcl_quads,
+                    shrink_ratio_of_width=shrink_ratio_of_width,
+                    expand_height_ratio=1.0 / tcl_ratio)
                 # generate tcl map
-                cv2.fillPoly(score_map, np.round(stcl_quads).astype(np.int32), 1.0)
+                cv2.fillPoly(score_map,
+                             np.round(stcl_quads).astype(np.int32), 1.0)
 
                 # generate tbo map
                 for idx, quad in enumerate(stcl_quads):
                     quad_mask = np.zeros((h, w), dtype=np.float32)
-                    quad_mask = cv2.fillPoly(quad_mask, np.round(quad[np.newaxis, :, :]).astype(np.int32), 1.0)
-                    tbo_map = self.gen_quad_tbo(poly_quads[quad_index[idx]], quad_mask, tbo_map)
+                    quad_mask = cv2.fillPoly(
+                        quad_mask,
+                        np.round(quad[np.newaxis, :, :]).astype(np.int32), 1.0)
+                    tbo_map = self.gen_quad_tbo(poly_quads[quad_index[idx]],
+                                                quad_mask, tbo_map)
         return score_map, tbo_map, training_mask
 
-    def generate_tvo_and_tco(self, hw, polys, tags, tcl_ratio=0.3, ds_ratio=0.25):
+    def generate_tvo_and_tco(self,
+                             hw,
+                             polys,
+                             tags,
+                             tcl_ratio=0.3,
+                             ds_ratio=0.25):
         """
         Generate tcl map, tvo map and tbo map.
         """
@@ -297,35 +344,44 @@ class SASTProcessTrain(object):
 
             # generate min_area_quad
             min_area_quad, center_point = self.gen_min_area_quad_from_poly(poly)
-            min_area_quad_h = 0.5 * (np.linalg.norm(min_area_quad[0] - min_area_quad[3]) +
-                                    np.linalg.norm(min_area_quad[1] - min_area_quad[2]))
-            min_area_quad_w = 0.5 * (np.linalg.norm(min_area_quad[0] - min_area_quad[1]) +
-                                    np.linalg.norm(min_area_quad[2] - min_area_quad[3]))
+            min_area_quad_h = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[3]) +
+                np.linalg.norm(min_area_quad[1] - min_area_quad[2]))
+            min_area_quad_w = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[1]) +
+                np.linalg.norm(min_area_quad[2] - min_area_quad[3]))
 
             # generate tcl map and text, 128 * 128
             tcl_poly = self.poly2tcl(poly, tcl_ratio)
 
             # generate poly_tv_xy_map
             for idx in range(4):
-                cv2.fillPoly(poly_tv_xy_map[2 * idx],
-                            np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
-                            float(min(max(min_area_quad[idx, 0], 0), w)))
-                cv2.fillPoly(poly_tv_xy_map[2 * idx + 1],
-                            np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
-                            float(min(max(min_area_quad[idx, 1], 0), h)))
+                cv2.fillPoly(
+                    poly_tv_xy_map[2 * idx],
+                    np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                    float(min(max(min_area_quad[idx, 0], 0), w)))
+                cv2.fillPoly(
+                    poly_tv_xy_map[2 * idx + 1],
+                    np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                    float(min(max(min_area_quad[idx, 1], 0), h)))
 
             # generate poly_tc_xy_map
             for idx in range(2):
-                cv2.fillPoly(poly_tc_xy_map[idx],
-                            np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32), float(center_point[idx]))
+                cv2.fillPoly(
+                    poly_tc_xy_map[idx],
+                    np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                    float(center_point[idx]))
 
             # generate poly_short_edge_map
-            cv2.fillPoly(poly_short_edge_map,
-                        np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
-                        float(max(min(min_area_quad_h, min_area_quad_w), 1.0)))
+            cv2.fillPoly(
+                poly_short_edge_map,
+                np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                float(max(min(min_area_quad_h, min_area_quad_w), 1.0)))
 
             # generate poly_mask and training_mask
-            cv2.fillPoly(poly_mask, np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32), 1)
+            cv2.fillPoly(poly_mask,
+                         np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                         1)
 
         tvo_map *= poly_mask
         tvo_map[:8] -= poly_tv_xy_map
@@ -356,7 +412,8 @@ class SASTProcessTrain(object):
         elif point_num > 4:
             vector_1 = poly[0] - poly[1]
             vector_2 = poly[1] - poly[2]
-            cos_theta = np.dot(vector_1, vector_2) / (np.linalg.norm(vector_1) * np.linalg.norm(vector_2) + 1e-6)
+            cos_theta = np.dot(vector_1, vector_2) / (
+                np.linalg.norm(vector_1) * np.linalg.norm(vector_2) + 1e-6)
             theta = np.arccos(np.round(cos_theta, decimals=4))
 
             if abs(theta) > (70 / 180 * math.pi):
@@ -374,7 +431,8 @@ class SASTProcessTrain(object):
             min_area_quad = poly
             center_point = np.sum(poly, axis=0) / 4
         else:
-            rect = cv2.minAreaRect(poly.astype(np.int32))  # (center (x,y), (width, height), angle of rotation)
+            rect = cv2.minAreaRect(poly.astype(
+                np.int32))  # (center (x,y), (width, height), angle of rotation)
             center_point = rect[0]
             box = np.array(cv2.boxPoints(rect))
 
@@ -394,16 +452,23 @@ class SASTProcessTrain(object):
 
         return min_area_quad, center_point
 
-    def shrink_quad_along_width(self, quad, begin_width_ratio=0., end_width_ratio=1.):
+    def shrink_quad_along_width(self,
+                                quad,
+                                begin_width_ratio=0.,
+                                end_width_ratio=1.):
         """
         Generate shrink_quad_along_width.
         """
-        ratio_pair = np.array([[begin_width_ratio], [end_width_ratio]], dtype=np.float32)
+        ratio_pair = np.array(
+            [[begin_width_ratio], [end_width_ratio]], dtype=np.float32)
         p0_1 = quad[0] + (quad[1] - quad[0]) * ratio_pair
         p3_2 = quad[3] + (quad[2] - quad[3]) * ratio_pair
         return np.array([p0_1[0], p0_1[1], p3_2[1], p3_2[0]])
 
-    def shrink_poly_along_width(self, quads, shrink_ratio_of_width, expand_height_ratio=1.0):
+    def shrink_poly_along_width(self,
+                                quads,
+                                shrink_ratio_of_width,
+                                expand_height_ratio=1.0):
         """
         shrink poly with given length.
         """
@@ -421,22 +486,28 @@ class SASTProcessTrain(object):
             upper_edge_list.append(upper_edge_len)
 
         # length of left edge and right edge.
-        left_length = np.linalg.norm(quads[0][0] - quads[0][3]) * expand_height_ratio
-        right_length = np.linalg.norm(quads[-1][1] - quads[-1][2]) * expand_height_ratio
+        left_length = np.linalg.norm(quads[0][0] - quads[0][
+            3]) * expand_height_ratio
+        right_length = np.linalg.norm(quads[-1][1] - quads[-1][
+            2]) * expand_height_ratio
 
-        shrink_length = min(left_length, right_length, sum(upper_edge_list)) * shrink_ratio_of_width
+        shrink_length = min(left_length, right_length,
+                            sum(upper_edge_list)) * shrink_ratio_of_width
         # shrinking length
         upper_len_left = shrink_length
         upper_len_right = sum(upper_edge_list) - shrink_length
 
         left_idx, left_ratio = get_cut_info(upper_edge_list, upper_len_left)
-        left_quad = self.shrink_quad_along_width(quads[left_idx], begin_width_ratio=left_ratio, end_width_ratio=1)
+        left_quad = self.shrink_quad_along_width(
+            quads[left_idx], begin_width_ratio=left_ratio, end_width_ratio=1)
         right_idx, right_ratio = get_cut_info(upper_edge_list, upper_len_right)
-        right_quad = self.shrink_quad_along_width(quads[right_idx], begin_width_ratio=0, end_width_ratio=right_ratio)
-        
+        right_quad = self.shrink_quad_along_width(
+            quads[right_idx], begin_width_ratio=0, end_width_ratio=right_ratio)
+
         out_quad_list = []
         if left_idx == right_idx:
-            out_quad_list.append([left_quad[0], right_quad[1], right_quad[2], left_quad[3]])
+            out_quad_list.append(
+                [left_quad[0], right_quad[1], right_quad[2], left_quad[3]])
         else:
             out_quad_list.append(left_quad)
             for idx in range(left_idx + 1, right_idx):
@@ -500,7 +571,8 @@ class SASTProcessTrain(object):
         """
         Generate center line by poly clock-wise point. (4, 2)
         """
-        ratio_pair = np.array([[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
+        ratio_pair = np.array(
+            [[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
         p0_3 = poly[0] + (poly[3] - poly[0]) * ratio_pair
         p1_2 = poly[1] + (poly[2] - poly[1]) * ratio_pair
         return np.array([p0_3[0], p1_2[0], p1_2[1], p0_3[1]])
@@ -509,12 +581,14 @@ class SASTProcessTrain(object):
         """
         Generate center line by poly clock-wise point.
         """
-        ratio_pair = np.array([[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
+        ratio_pair = np.array(
+            [[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
         tcl_poly = np.zeros_like(poly)
         point_num = poly.shape[0]
 
         for idx in range(point_num // 2):
-            point_pair = poly[idx] + (poly[point_num - 1 - idx] - poly[idx]) * ratio_pair
+            point_pair = poly[idx] + (poly[point_num - 1 - idx] - poly[idx]
+                                      ) * ratio_pair
             tcl_poly[idx] = point_pair[0]
             tcl_poly[point_num - 1 - idx] = point_pair[1]
         return tcl_poly
@@ -527,8 +601,10 @@ class SASTProcessTrain(object):
         up_line = self.line_cross_two_point(quad[0], quad[1])
         lower_line = self.line_cross_two_point(quad[3], quad[2])
 
-        quad_h = 0.5 * (np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[1] - quad[2]))
-        quad_w = 0.5 * (np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[2] - quad[3]))
+        quad_h = 0.5 * (np.linalg.norm(quad[0] - quad[3]) +
+                        np.linalg.norm(quad[1] - quad[2]))
+        quad_w = 0.5 * (np.linalg.norm(quad[0] - quad[1]) +
+                        np.linalg.norm(quad[2] - quad[3]))
 
         # average angle of left and right line.
         angle = self.average_angle(quad)
@@ -565,7 +641,8 @@ class SASTProcessTrain(object):
         quad_num = point_num // 2 - 1
         for idx in range(quad_num):
             # reshape and adjust to clock-wise
-            quad_list.append((np.array(point_pair_list)[[idx, idx + 1]]).reshape(4, 2)[[0, 2, 3, 1]])
+            quad_list.append((np.array(point_pair_list)[[idx, idx + 1]]
+                              ).reshape(4, 2)[[0, 2, 3, 1]])
 
         return np.array(quad_list)
 
@@ -579,7 +656,8 @@ class SASTProcessTrain(object):
             return None
 
         h, w, _ = im.shape
-        text_polys, text_tags, hv_tags = self.check_and_validate_polys(text_polys, text_tags, (h, w))
+        text_polys, text_tags, hv_tags = self.check_and_validate_polys(
+            text_polys, text_tags, (h, w))
 
         if text_polys.shape[0] == 0:
             return None
@@ -591,7 +669,7 @@ class SASTProcessTrain(object):
         if np.random.rand() < 0.5:
             asp_scale = 1.0 / asp_scale
         asp_scale = math.sqrt(asp_scale)
-        
+
         asp_wx = asp_scale
         asp_hy = 1.0 / asp_scale
         im = cv2.resize(im, dsize=None, fx=asp_wx, fy=asp_hy)
@@ -610,7 +688,7 @@ class SASTProcessTrain(object):
         #no background
         im, text_polys, text_tags, hv_tags = self.crop_area(im, \
             text_polys, text_tags, hv_tags, crop_background=False)
-            
+
         if text_polys.shape[0] == 0:
             return None
         #continue for all ignore case
@@ -621,17 +699,18 @@ class SASTProcessTrain(object):
             return None
         #resize image
         std_ratio = float(self.input_size) / max(new_w, new_h)
-        rand_scales = np.array([0.25, 0.375, 0.5, 0.625, 0.75, 0.875, 1.0, 1.0, 1.0, 1.0, 1.0])
+        rand_scales = np.array(
+            [0.25, 0.375, 0.5, 0.625, 0.75, 0.875, 1.0, 1.0, 1.0, 1.0, 1.0])
         rz_scale = std_ratio * np.random.choice(rand_scales)
         im = cv2.resize(im, dsize=None, fx=rz_scale, fy=rz_scale)
         text_polys[:, :, 0] *= rz_scale
         text_polys[:, :, 1] *= rz_scale
-        
+
         #add gaussian blur
         if np.random.rand() < 0.1 * 0.5:
             ks = np.random.permutation(5)[0] + 1
-            ks = int(ks/2)*2 + 1
-            im =  cv2.GaussianBlur(im, ksize=(ks, ks), sigmaX=0, sigmaY=0)
+            ks = int(ks / 2) * 2 + 1
+            im = cv2.GaussianBlur(im, ksize=(ks, ks), sigmaX=0, sigmaY=0)
         #add brighter
         if np.random.rand() < 0.1 * 0.5:
             im = im * (1.0 + np.random.rand() * 0.5)
@@ -640,13 +719,14 @@ class SASTProcessTrain(object):
         if np.random.rand() < 0.1 * 0.5:
             im = im * (1.0 - np.random.rand() * 0.5)
             im = np.clip(im, 0.0, 255.0)
-        
+
         # Padding the im to [input_size, input_size]
         new_h, new_w, _ = im.shape
         if min(new_w, new_h) < self.input_size * 0.5:
             return None
 
-        im_padded = np.ones((self.input_size, self.input_size, 3), dtype=np.float32)
+        im_padded = np.ones(
+            (self.input_size, self.input_size, 3), dtype=np.float32)
         im_padded[:, :, 2] = 0.485 * 255
         im_padded[:, :, 1] = 0.456 * 255
         im_padded[:, :, 0] = 0.406 * 255
@@ -661,24 +741,29 @@ class SASTProcessTrain(object):
             sw = int(np.random.rand() * del_w)
 
         # Padding
-        im_padded[sh: sh + new_h, sw: sw + new_w, :] = im.copy()
+        im_padded[sh:sh + new_h, sw:sw + new_w, :] = im.copy()
         text_polys[:, :, 0] += sw
         text_polys[:, :, 1] += sh
 
-        score_map, border_map, training_mask = self.generate_tcl_label((self.input_size, self.input_size), 
-                            text_polys, text_tags, 0.25)
-        
+        score_map, border_map, training_mask = self.generate_tcl_label(
+            (self.input_size, self.input_size), text_polys, text_tags, 0.25)
+
         # SAST head
-        tvo_map, tco_map = self.generate_tvo_and_tco((self.input_size, self.input_size), text_polys, text_tags,  tcl_ratio=0.3, ds_ratio=0.25)
+        tvo_map, tco_map = self.generate_tvo_and_tco(
+            (self.input_size, self.input_size),
+            text_polys,
+            text_tags,
+            tcl_ratio=0.3,
+            ds_ratio=0.25)
         # print("test--------tvo_map shape:", tvo_map.shape)
 
         im_padded[:, :, 2] -= 0.485 * 255
         im_padded[:, :, 1] -= 0.456 * 255
         im_padded[:, :, 0] -= 0.406 * 255
-        im_padded[:, :, 2] /= (255.0 * 0.229) 
-        im_padded[:, :, 1] /= (255.0 * 0.224) 
-        im_padded[:, :, 0] /= (255.0 * 0.225) 
-        im_padded = im_padded.transpose((2, 0, 1))        
+        im_padded[:, :, 2] /= (255.0 * 0.229)
+        im_padded[:, :, 1] /= (255.0 * 0.224)
+        im_padded[:, :, 0] /= (255.0 * 0.225)
+        im_padded = im_padded.transpose((2, 0, 1))
 
         data['image'] = im_padded[::-1, :, :]
         data['score_map'] = score_map[np.newaxis, :, :]
@@ -686,4 +771,4 @@ class SASTProcessTrain(object):
         data['training_mask'] = training_mask[np.newaxis, :, :]
         data['tvo_map'] = tvo_map.transpose((2, 0, 1))
         data['tco_map'] = tco_map.transpose((2, 0, 1))
-        return data
+        return data
\ No newline at end of file

ppocr/data/lmdb_dataset.py

@@ -21,7 +21,7 @@ from .imaug import transform, create_operators
 
 
 class LMDBDateSet(Dataset):
-    def __init__(self, config, mode, logger):
+    def __init__(self, config, mode, logger, seed=None):
         super(LMDBDateSet, self).__init__()
 
         global_config = config['Global']

ppocr/data/simple_dataset.py

@@ -20,7 +20,7 @@ from .imaug import transform, create_operators
 
 
 class SimpleDataSet(Dataset):
-    def __init__(self, config, mode, logger):
+    def __init__(self, config, mode, logger, seed=None):
         super(SimpleDataSet, self).__init__()
         self.logger = logger
 
@@ -41,6 +41,7 @@ class SimpleDataSet(Dataset):
         self.data_dir = dataset_config['data_dir']
         self.do_shuffle = loader_config['shuffle']
 
+        self.seed = seed
         logger.info("Initialize indexs of datasets:%s" % label_file_list)
         self.data_lines = self.get_image_info_list(label_file_list, ratio_list)
         self.data_idx_order_list = list(range(len(self.data_lines)))
@@ -55,6 +56,7 @@ class SimpleDataSet(Dataset):
         for idx, file in enumerate(file_list):
             with open(file, "rb") as f:
                 lines = f.readlines()
+                random.seed(self.seed)
                 lines = random.sample(lines,
                                       round(len(lines) * ratio_list[idx]))
                 data_lines.extend(lines)
@@ -62,6 +64,7 @@ class SimpleDataSet(Dataset):
 
     def shuffle_data_random(self):
         if self.do_shuffle:
+            random.seed(self.seed)
             random.shuffle(self.data_lines)
         return
 

ppocr/modeling/transforms/tps.py

@@ -213,16 +213,14 @@ class GridGenerator(nn.Layer):
 
     def build_P_paddle(self, I_r_size):
         I_r_height, I_r_width = I_r_size
-        I_r_grid_x = paddle.divide(
-            paddle.arange(
-                -I_r_width, I_r_width, 2, dtype='float64') + 1.0,
-            paddle.to_tensor(
-                I_r_width, dtype='float64'))
-        I_r_grid_y = paddle.divide(
-            paddle.arange(
-                -I_r_height, I_r_height, 2, dtype='float64') + 1.0,
-            paddle.to_tensor(
-                I_r_height, dtype='float64'))  # self.I_r_height
+        I_r_grid_x = (paddle.arange(
+            -I_r_width, I_r_width, 2, dtype='float64') + 1.0
+                      ) / paddle.to_tensor(np.array([I_r_width]))
+
+        I_r_grid_y = (paddle.arange(
+            -I_r_height, I_r_height, 2, dtype='float64') + 1.0
+                      ) / paddle.to_tensor(np.array([I_r_height]))
+
         # P: self.I_r_width x self.I_r_height x 2
         P = paddle.stack(paddle.meshgrid(I_r_grid_x, I_r_grid_y), axis=2)
         P = paddle.transpose(P, perm=[1, 0, 2])

ppocr/postprocess/rec_postprocess.py

@@ -109,7 +109,7 @@ class CTCLabelDecode(BaseRecLabelDecode):
 
         preds_idx = preds.argmax(axis=2)
         preds_prob = preds.max(axis=2)
-        text = self.decode(preds_idx, preds_prob)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True)
         if label is None:
             return text
         label = self.decode(label)

tools/program.py

@@ -182,8 +182,8 @@ def train(config,
         start_epoch = 1
 
     for epoch in range(start_epoch, epoch_num + 1):
-        if epoch > 0:
-            train_dataloader = build_dataloader(config, 'Train', device, logger)
+        train_dataloader = build_dataloader(
+            config, 'Train', device, logger, seed=epoch)
         train_batch_cost = 0.0
         train_reader_cost = 0.0
         batch_sum = 0