Merge branch 'PaddlePaddle:release/2.5' into onnx

PPOCRLabel/PPOCRLabel.py

@@ -1149,7 +1149,10 @@ class MainWindow(QMainWindow):
         for box in self.result_dic:
             trans_dic = {"label": box[1][0], "points": box[0], "difficult": False}
             if self.kie_mode:
-                trans_dic.update({"key_cls": "None"})
+                if len(box) == 3:
+                    trans_dic.update({"key_cls": box[2]})
+                else:
+                    trans_dic.update({"key_cls": "None"})
             if trans_dic["label"] == "" and mode == 'Auto':
                 continue
             shapes.append(trans_dic)
@@ -2047,6 +2050,7 @@ class MainWindow(QMainWindow):
             rec_flag = 0
             for shape in self.canvas.shapes:
                 box = [[int(p.x()), int(p.y())] for p in shape.points]
+                kie_cls = shape.key_cls
 
                 if len(box) > 4:
                     box = self.gen_quad_from_poly(np.array(box))
@@ -2062,17 +2066,27 @@ class MainWindow(QMainWindow):
                     if shape.line_color == DEFAULT_LOCK_COLOR:
                         shape.label = result[0][0]
                         result.insert(0, box)
+                        if self.kie_mode:
+                            result.append(kie_cls)
                         self.result_dic_locked.append(result)
                     else:
                         result.insert(0, box)
+                        if self.kie_mode:
+                            result.append(kie_cls)
                         self.result_dic.append(result)
                 else:
                     print('Can not recognise the box')
                     if shape.line_color == DEFAULT_LOCK_COLOR:
                         shape.label = result[0][0]
-                        self.result_dic_locked.append([box, (self.noLabelText, 0)])
+                        if self.kie_mode:
+                            self.result_dic_locked.append([box, (self.noLabelText, 0), kie_cls])
+                        else:
+                            self.result_dic_locked.append([box, (self.noLabelText, 0)])
                     else:
-                        self.result_dic.append([box, (self.noLabelText, 0)])
+                        if self.kie_mode:
+                            self.result_dic.append([box, (self.noLabelText, 0), kie_cls])
+                        else:
+                            self.result_dic.append([box, (self.noLabelText, 0)])
                 try:
                     if self.noLabelText == shape.label or result[1][0] == shape.label:
                         print('label no change')
@@ -2322,13 +2336,6 @@ class MainWindow(QMainWindow):
                     else:
                         labeldict[file] = []
 
-        # if len(labeldict) != len(csv_paths):
-        #     msg = 'ERROR, box label and excel label are not in the same number\n' + \
-        #           'box label: ' + str(len(labeldict)) + '\n' + \
-        #           'excel label: ' + str(len(csv_paths)) + '\n' + \
-        #           'Please check the label.txt and tableRec_excel_output\n'
-        #     QMessageBox.information(self, "Information", msg)
-        #     return
         train_split, val_split, test_split = partitionDialog.getDataPartition()
         # check validate
         if train_split + val_split + test_split > 100:
@@ -2351,14 +2358,8 @@ class MainWindow(QMainWindow):
             filename, _ = os.path.splitext(os.path.basename(image_path))
             csv_path = os.path.join(TableRec_excel_dir, filename + '.xlsx')
             if not os.path.exists(csv_path):
-                msg = 'ERROR, Can not find ' + csv_path
-                QMessageBox.information(self, "Information", msg)
-                return
+                continue
 
-            # read xlsx file, convert to HTML
-            # xd = pd.ExcelFile(csv_path)
-            # df = xd.parse()
-            # structure = df.to_html(index = False)
             excel = xlrd.open_workbook(csv_path)
             sheet0 = excel.sheet_by_index(0)  # only sheet 0
             merged_cells = sheet0.merged_cells # (0,1,1,3) start row, end row, start col, end col
@@ -2370,7 +2371,6 @@ class MainWindow(QMainWindow):
 
             token_list = convert_token(html_list)
 
-
             # load box annotations
             cells = []
             for anno in labeldict[image_path]:

PPOCRLabel/README.md

@@ -206,7 +206,7 @@ the table and pop up Excel at the same time.
 - Model language switching: Changing the built-in model language is supportable by clicking "PaddleOCR"-"Choose OCR Model" in the menu bar. Currently supported languages​include French, German, Korean, and Japanese.
   For specific model download links, please refer to [PaddleOCR Model List](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_en/models_list_en.md#multilingual-recognition-modelupdating)
 
-- **Custom Model**: If users want to replace the built-in model with their own inference model, they can follow the [Custom Model Code Usage](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.3/doc/doc_en/whl_en.md#31-use-by-code) by modifying PPOCRLabel.py for [Instantiation of PaddleOCR class](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/PPOCRLabel/PPOCRLabel.py#L86) :
+- **Custom Model**: If users want to replace the built-in model with their own inference model, they can follow the [Custom Model Code Usage](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.3/doc/doc_en/whl_en.md#31-use-by-code) by modifying PPOCRLabel.py for [Instantiation of PaddleOCR class](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/PPOCRLabel/PPOCRLabel.py#L97) :
 
   add parameter `det_model_dir`  in `self.ocr = PaddleOCR(use_pdserving=False, use_angle_cls=True, det=True, cls=True, use_gpu=gpu, lang=lang) `
 

PPOCRLabel/libs/utils.py

@@ -221,10 +221,10 @@ def convert_token(html_list):
                 token_list.append("<td")
                 if 'colspan' in col:
                     _, n = col.split('colspan=')
-                    token_list.append(" colspan=\"{}\"".format(n))
+                    token_list.append(" colspan=\"{}\"".format(n[0]))
                 if 'rowspan' in col:
                     _, n = col.split('rowspan=')
-                    token_list.append(" rowspan=\"{}\"".format(n))
+                    token_list.append(" rowspan=\"{}\"".format(n[0]))
                 token_list.extend([">", "</td>"])
         token_list.append("</tr>")
     token_list.append("</tbody>")

PPOCRLabel/resources/strings/strings-en.properties

@@ -112,4 +112,4 @@ keyDialogTip=Enter object label
 keyChange=Change Box Key
 TableRecognition=Table Recognition
 cellreRecognition=Cell Re-Recognition
-exportJSON=export JSON(PubTabNet)
+exportJSON=Export Excel Label(PubTabNet)

PPOCRLabel/resources/strings/strings-zh-CN.properties

@@ -84,7 +84,7 @@ mhelp=帮助
 iconList=缩略图
 detectionBoxposition=检测框位置
 recognitionResult=识别结果
-creatPolygon=多边形标注
+creatPolygon=多点标注
 drawSquares=正方形标注
 rotateLeft=图片左旋转90度
 rotateRight=图片右旋转90度

deploy/pdserving/README.md

@@ -163,43 +163,41 @@ The recognition model is the same.
 
     The predicted performance data will be automatically written into the `PipelineServingLogs/pipeline.tracer` file.
 
-    Tested on 200 real pictures, and limited the detection long side to 960. The average QPS on T4 GPU can reach around 23:
+    Tested on 200 real pictures, and limited the detection long side to 960. The average QPS on T4 GPU can reach around 62.0:
 
     ```
-
-    2021-05-13 03:42:36,895 ==================== TRACER ======================
-    2021-05-13 03:42:36,975 Op(rec):
-    2021-05-13 03:42:36,976         in[14.472382882882883 ms]
-    2021-05-13 03:42:36,976         prep[9.556855855855856 ms]
-    2021-05-13 03:42:36,976         midp[59.921905405405404 ms]
-    2021-05-13 03:42:36,976         postp[15.345945945945946 ms]
-    2021-05-13 03:42:36,976         out[1.9921216216216215 ms]
-    2021-05-13 03:42:36,976         idle[0.16254943864471572]
-    2021-05-13 03:42:36,976 Op(det):
-    2021-05-13 03:42:36,976         in[315.4468035714286 ms]
-    2021-05-13 03:42:36,976         prep[69.5980625 ms]
-    2021-05-13 03:42:36,976         midp[18.989535714285715 ms]
-    2021-05-13 03:42:36,976         postp[18.857803571428573 ms]
-    2021-05-13 03:42:36,977         out[3.1337544642857145 ms]
-    2021-05-13 03:42:36,977         idle[0.7477961159203756]
-    2021-05-13 03:42:36,977 DAGExecutor:
-    2021-05-13 03:42:36,977         Query count[224]
-    2021-05-13 03:42:36,977         QPS[22.4 q/s]
-    2021-05-13 03:42:36,977         Succ[0.9910714285714286]
-    2021-05-13 03:42:36,977         Error req[169, 170]
-    2021-05-13 03:42:36,977         Latency:
-    2021-05-13 03:42:36,977                 ave[535.1678348214285 ms]
-    2021-05-13 03:42:36,977                 .50[172.651 ms]
-    2021-05-13 03:42:36,977                 .60[187.904 ms]
-    2021-05-13 03:42:36,977                 .70[245.675 ms]
-    2021-05-13 03:42:36,977                 .80[526.684 ms]
-    2021-05-13 03:42:36,977                 .90[854.596 ms]
-    2021-05-13 03:42:36,977                 .95[1722.728 ms]
-    2021-05-13 03:42:36,977                 .99[3990.292 ms]
-    2021-05-13 03:42:36,978 Channel (server worker num[10]):
-    2021-05-13 03:42:36,978         chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
-    2021-05-13 03:42:36,979         chl1(In: ['det'], Out: ['rec']) size[6/0]
-    2021-05-13 03:42:36,979         chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
+    2022-05-12 03:56:46,461 ==================== TRACER ======================
+    2022-05-12 03:56:46,860 Op(det):
+    2022-05-12 03:56:46,860         in[80.32286641221374 ms]
+    2022-05-12 03:56:46,860         prep[74.27364885496183 ms]
+    2022-05-12 03:56:46,860         midp[33.41587786259542 ms]
+    2022-05-12 03:56:46,860         postp[20.935980916030534 ms]
+    2022-05-12 03:56:46,860         out[1.551145038167939 ms]
+    2022-05-12 03:56:46,860         idle[0.3889510617728378]
+    2022-05-12 03:56:46,860 Op(rec):
+    2022-05-12 03:56:46,860         in[15.46498846153846 ms]
+    2022-05-12 03:56:46,861         prep[22.565715384615384 ms]
+    2022-05-12 03:56:46,861         midp[91.42518076923076 ms]
+    2022-05-12 03:56:46,861         postp[11.678453846153847 ms]
+    2022-05-12 03:56:46,861         out[1.1200576923076924 ms]
+    2022-05-12 03:56:46,861         idle[0.11658723106110291]
+    2022-05-12 03:56:46,862 DAGExecutor:
+    2022-05-12 03:56:46,862         Query count[620]
+    2022-05-12 03:56:46,862         QPS[62.0 q/s]
+    2022-05-12 03:56:46,862         Succ[0.4193548387096774]
+    2022-05-12 03:56:46,862         Latency:
+    2022-05-12 03:56:46,863                 ave[165.54603709677417 ms]
+    2022-05-12 03:56:46,863                 .50[77.863 ms]
+    2022-05-12 03:56:46,863                 .60[158.414 ms]
+    2022-05-12 03:56:46,863                 .70[237.28 ms]
+    2022-05-12 03:56:46,863                 .80[316.022 ms]
+    2022-05-12 03:56:46,863                 .90[424.416 ms]
+    2022-05-12 03:56:46,863                 .95[515.566 ms]
+    2022-05-12 03:56:46,863                 .99[762.256 ms]
+    2022-05-12 03:56:46,863 Channel (server worker num[10]):
+    2022-05-12 03:56:46,864         chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
+    2022-05-12 03:56:46,864         chl1(In: ['det'], Out: ['rec']) size[2/0]
+    2022-05-12 03:56:46,865         chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
     ```
 
 <a name="C++"></a>

deploy/pdserving/README_CN.md

@@ -162,42 +162,41 @@ python3 -m paddle_serving_client.convert --dirname ./ch_PP-OCRv3_rec_infer/ \
 
     预测性能数据会被自动写入 `PipelineServingLogs/pipeline.tracer` 文件中。
 
-    在200张真实图片上测试，把检测长边限制为960。T4 GPU 上 QPS 均值可达到23左右：
+    在200张真实图片上测试，把检测长边限制为960。T4 GPU 上 QPS 均值可达到62左右：
 
     ```
-    2021-05-13 03:42:36,895 ==================== TRACER ======================
-    2021-05-13 03:42:36,975 Op(rec):
-    2021-05-13 03:42:36,976         in[14.472382882882883 ms]
-    2021-05-13 03:42:36,976         prep[9.556855855855856 ms]
-    2021-05-13 03:42:36,976         midp[59.921905405405404 ms]
-    2021-05-13 03:42:36,976         postp[15.345945945945946 ms]
-    2021-05-13 03:42:36,976         out[1.9921216216216215 ms]
-    2021-05-13 03:42:36,976         idle[0.16254943864471572]
-    2021-05-13 03:42:36,976 Op(det):
-    2021-05-13 03:42:36,976         in[315.4468035714286 ms]
-    2021-05-13 03:42:36,976         prep[69.5980625 ms]
-    2021-05-13 03:42:36,976         midp[18.989535714285715 ms]
-    2021-05-13 03:42:36,976         postp[18.857803571428573 ms]
-    2021-05-13 03:42:36,977         out[3.1337544642857145 ms]
-    2021-05-13 03:42:36,977         idle[0.7477961159203756]
-    2021-05-13 03:42:36,977 DAGExecutor:
-    2021-05-13 03:42:36,977         Query count[224]
-    2021-05-13 03:42:36,977         QPS[22.4 q/s]
-    2021-05-13 03:42:36,977         Succ[0.9910714285714286]
-    2021-05-13 03:42:36,977         Error req[169, 170]
-    2021-05-13 03:42:36,977         Latency:
-    2021-05-13 03:42:36,977                 ave[535.1678348214285 ms]
-    2021-05-13 03:42:36,977                 .50[172.651 ms]
-    2021-05-13 03:42:36,977                 .60[187.904 ms]
-    2021-05-13 03:42:36,977                 .70[245.675 ms]
-    2021-05-13 03:42:36,977                 .80[526.684 ms]
-    2021-05-13 03:42:36,977                 .90[854.596 ms]
-    2021-05-13 03:42:36,977                 .95[1722.728 ms]
-    2021-05-13 03:42:36,977                 .99[3990.292 ms]
-    2021-05-13 03:42:36,978 Channel (server worker num[10]):
-    2021-05-13 03:42:36,978         chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
-    2021-05-13 03:42:36,979         chl1(In: ['det'], Out: ['rec']) size[6/0]
-    2021-05-13 03:42:36,979         chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
+    2022-05-12 03:56:46,461 ==================== TRACER ======================
+    2022-05-12 03:56:46,860 Op(det):
+    2022-05-12 03:56:46,860         in[80.32286641221374 ms]
+    2022-05-12 03:56:46,860         prep[74.27364885496183 ms]
+    2022-05-12 03:56:46,860         midp[33.41587786259542 ms]
+    2022-05-12 03:56:46,860         postp[20.935980916030534 ms]
+    2022-05-12 03:56:46,860         out[1.551145038167939 ms]
+    2022-05-12 03:56:46,860         idle[0.3889510617728378]
+    2022-05-12 03:56:46,860 Op(rec):
+    2022-05-12 03:56:46,860         in[15.46498846153846 ms]
+    2022-05-12 03:56:46,861         prep[22.565715384615384 ms]
+    2022-05-12 03:56:46,861         midp[91.42518076923076 ms]
+    2022-05-12 03:56:46,861         postp[11.678453846153847 ms]
+    2022-05-12 03:56:46,861         out[1.1200576923076924 ms]
+    2022-05-12 03:56:46,861         idle[0.11658723106110291]
+    2022-05-12 03:56:46,862 DAGExecutor:
+    2022-05-12 03:56:46,862         Query count[620]
+    2022-05-12 03:56:46,862         QPS[62.0 q/s]
+    2022-05-12 03:56:46,862         Succ[0.4193548387096774]
+    2022-05-12 03:56:46,862         Latency:
+    2022-05-12 03:56:46,863                 ave[165.54603709677417 ms]
+    2022-05-12 03:56:46,863                 .50[77.863 ms]
+    2022-05-12 03:56:46,863                 .60[158.414 ms]
+    2022-05-12 03:56:46,863                 .70[237.28 ms]
+    2022-05-12 03:56:46,863                 .80[316.022 ms]
+    2022-05-12 03:56:46,863                 .90[424.416 ms]
+    2022-05-12 03:56:46,863                 .95[515.566 ms]
+    2022-05-12 03:56:46,863                 .99[762.256 ms]
+    2022-05-12 03:56:46,863 Channel (server worker num[10]):
+    2022-05-12 03:56:46,864         chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
+    2022-05-12 03:56:46,864         chl1(In: ['det'], Out: ['rec']) size[2/0]
+    2022-05-12 03:56:46,865         chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
     ```
 
 <a name="C++"></a>

doc/doc_ch/knowledge_distillation.md

@@ -293,8 +293,8 @@ Loss:
 
 以上述配置为例，最终蒸馏训练的损失函数包含下面5个部分。
 
-- `Student`和`Teacher`最终输出(`head_out`)的CTC分支与gt的CTC loss，权重为1。在这里因为2个子网络都需要更新参数，因此2者都需要计算与g的loss。
-- `Student`和`Teacher`最终输出(`head_out`)的SAR分支与gt的SAR loss，权重为1.0。在这里因为2个子网络都需要更新参数，因此2者都需要计算与g的loss。
+- `Student`和`Teacher`最终输出(`head_out`)的CTC分支与gt的CTC loss，权重为1。在这里因为2个子网络都需要更新参数，因此2者都需要计算与gt的loss。
+- `Student`和`Teacher`最终输出(`head_out`)的SAR分支与gt的SAR loss，权重为1.0。在这里因为2个子网络都需要更新参数，因此2者都需要计算与gt的loss。
 - `Student`和`Teacher`最终输出(`head_out`)的CTC分支之间的DML loss，权重为1。
 - `Student`和`Teacher`最终输出(`head_out`)的SAR分支之间的DML loss，权重为0.5。
 - `Student`和`Teacher`的骨干网络输出(`backbone_out`)之间的l2 loss，权重为1。
@@ -374,7 +374,7 @@ paddle.save(s_params, "ch_PP-OCRv3_rec_train/student.pdparams")
 <a name="22"></a>
 ### 2.2 检测配置文件解析
 
-检测模型蒸馏的配置文件在PaddleOCR/configs/det/ch_PP-OCRv3/目录下，包含两个个蒸馏配置文件：
+检测模型蒸馏的配置文件在PaddleOCR/configs/det/ch_PP-OCRv3/目录下，包含两个蒸馏配置文件：
 - ch_PP-OCRv3_det_cml.yml，采用cml蒸馏，采用一个大模型蒸馏两个小模型，且两个小模型互相学习的方法
 - ch_PP-OCRv3_det_dml.yml，采用DML的蒸馏，两个Student模型互蒸馏的方法
 
@@ -383,7 +383,7 @@ paddle.save(s_params, "ch_PP-OCRv3_rec_train/student.pdparams")
 
 知识蒸馏任务中，模型结构配置如下所示：
 
-```
+```yaml
 Architecture:
   name: DistillationModel          # 结构名称，蒸馏任务中，为DistillationModel，用于构建对应的结构
   algorithm: Distillation          # 算法名称
@@ -424,11 +424,11 @@ Architecture:
 
 ```
 
-如果是采用DML，即两个小模型互相学习的方法，上述配置文件里的Teacher网络结构需要设置为Student模型一样的配置，具体参考配置文件[ch_PP-OCRv3_det_dml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.4/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml)。
-
-下面介绍[ch_PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.4/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml)的配置文件参数：
+如果是采用DML，即两个小模型互相学习的方法，上述配置文件里的Teacher网络结构需要设置为Student模型一样的配置，具体参考配置文件[ch_PP-OCRv3_det_dml.yml](../../configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml)。
 
-```
+下面介绍[ch_PP-OCRv3_det_cml.yml](../../configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml)的配置文件参数：
+
+```yaml
 Architecture:
   name: DistillationModel  
   algorithm: Distillation

doc/doc_ch/models_list.md

@@ -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
 
 |模型类型|模型格式|简介|
 |--- | --- | --- |
-|推理模型|inference.pdmodel、inference.pdiparams|用于预测引擎推理，[详情](./inference.md)|
+|推理模型|inference.pdmodel、inference.pdiparams|用于预测引擎推理，[详情](./inference_ppocr.md)|
 |训练模型、预训练模型|\*.pdparams、\*.pdopt、\*.states |训练过程中保存的模型的参数、优化器状态和训练中间信息，多用于模型指标评估和恢复训练|
 |nb模型|\*.nb|经过飞桨Paddle-Lite工具优化后的模型，适用于移动端/IoT端等端侧部署场景（需使用飞桨Paddle Lite部署）。|
 
@@ -114,7 +114,7 @@ PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训
 | ka_PP-OCRv3_rec | ppocr/utils/dict/ka_dict.txt |卡纳达文识别|[ka_PP-OCRv3_rec.yml](../../configs/rec/PP-OCRv3/multi_language/ka_PP-OCRv3_rec.yml)|9.9M|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ka_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ka_PP-OCRv3_rec_train.tar) |
 | ta_PP-OCRv3_rec | ppocr/utils/dict/ta_dict.txt |泰米尔文识别|[ta_PP-OCRv3_rec.yml](../../configs/rec/PP-OCRv3/multi_language/ta_PP-OCRv3_rec.yml)|9.6M|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ta_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ta_PP-OCRv3_rec_train.tar) |
 | latin_PP-OCRv3_rec |  ppocr/utils/dict/latin_dict.txt | 拉丁文识别 |  [latin_PP-OCRv3_rec.yml](../../configs/rec/PP-OCRv3/multi_language/latin_PP-OCRv3_rec.yml) |9.7M|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/latin_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/latin_PP-OCRv3_rec_train.tar) |
-| arabic_PP-OCRv3_rec | ppocr/utils/dict/arabic_dict.txt | 阿拉伯字母 | [arabic_PP-OCRv3_rec.yml](../../configs/rec/PP-OCRv3/multi_language/rec_arabic_lite_train.yml) |9.6M|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/arabic_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/arabic_PP-OCRv3_rec_train.tar) |
+| arabic_PP-OCRv3_rec | ppocr/utils/dict/arabic_dict.txt | 阿拉伯字母 | [arabic_PP-OCRv3_rec.yml](../../configs/rec/PP-OCRv3/multi_language/arabic_PP-OCRv3_rec.yml) |9.6M|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/arabic_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/arabic_PP-OCRv3_rec_train.tar) |
 | cyrillic_PP-OCRv3_rec | ppocr/utils/dict/cyrillic_dict.txt | 斯拉夫字母 | [cyrillic_PP-OCRv3_rec.yml](../../configs/rec/PP-OCRv3/multi_language/cyrillic_PP-OCRv3_rec.yml) |9.6M|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/cyrillic_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/cyrillic_PP-OCRv3_rec_train.tar) |
 | devanagari_PP-OCRv3_rec | ppocr/utils/dict/devanagari_dict.txt |梵文字母 | [devanagari_PP-OCRv3_rec.yml](../../configs/rec/PP-OCRv3/multi_language/devanagari_PP-OCRv3_rec.yml) |9.9M|[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/devanagari_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/devanagari_PP-OCRv3_rec_train.tar) |
 

doc/doc_ch/multi_languages.md

@@ -238,10 +238,10 @@ python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs
 ## 4 预测部署
 
 除了安装whl包进行快速预测，ppocr 也提供了多种预测部署方式，如有需求可阅读相关文档：
-- [基于Python脚本预测引擎推理](./inference.md)
-- [基于C++预测引擎推理](../../deploy/cpp_infer/readme.md)
+- [基于Python脚本预测引擎推理](./inference_ppocr.md)
+- [基于C++预测引擎推理](../../deploy/cpp_infer/readme_ch.md)
 - [服务化部署](../../deploy/hubserving/readme.md)
-- [端侧部署](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/deploy/lite/readme.md)
+- [端侧部署](../../deploy/lite/readme_ch.md)
 - [Benchmark](./benchmark.md)
 
 

doc/doc_ch/serving_inference.md

-# 使用Paddle Serving预测推理
-
-阅读本文档之前，请先阅读文档 [基于Python预测引擎推理](./inference.md)
-
-同本地执行预测一样，我们需要保存一份可以用于Paddle Serving的模型。
-
-接下来首先介绍如何将训练的模型转换成Paddle Serving模型，然后将依次介绍文本检测、文本识别以及两者串联基于预测引擎推理。
-
-### 一、 准备环境
-我们先安装Paddle Serving相关组件
-我们推荐用户使用GPU来做Paddle Serving的OCR服务部署
-
-**CUDA版本：9.X/10.X**
-
-**CUDNN版本：7.X**
-
-**操作系统版本：Linux/Windows**
-
-**Python版本： 2.7/3.5/3.6/3.7**
-
-**Python操作指南：**
-
-目前Serving用于OCR的部分功能还在测试当中，因此在这里我们给出[Servnig latest package](https://github.com/PaddlePaddle/Serving/blob/develop/doc/Latest_Packages_CN.md)
-大家根据自己的环境选择需要安装的whl包即可，例如以Python 3.5为例，执行下列命令
-```
-#CPU/GPU版本选择一个
-#GPU版本服务端
-#CUDA 9
-python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post9-py3-none-any.whl
-#CUDA 10
-python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post10-py3-none-any.whl
-#CPU版本服务端
-python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server-0.0.0-py3-none-any.whl
-#客户端和App包使用以下链接（CPU，GPU通用）
-python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.0.0-cp36-none-any.whl https://paddle-serving.bj.bcebos.com/whl/paddle_serving_app-0.0.0-py3-none-any.whl
-```
-
-## 二、训练模型转Serving模型
-
-在前序文档 [基于Python预测引擎推理](./inference.md) 中，我们提供了如何把训练的checkpoint转换成Paddle模型。Paddle模型通常由一个文件夹构成，内含模型结构描述文件`model`和模型参数文件`params`。Serving模型由两个文件夹构成，用于存放客户端和服务端的配置。
-
-我们以`ch_rec_r34_vd_crnn`模型作为例子，下载链接在：
-
-```
-wget --no-check-certificate https://paddleocr.bj.bcebos.com/ch_models/ch_rec_r34_vd_crnn_infer.tar
-tar xf ch_rec_r34_vd_crnn_infer.tar
-```
-因此我们按照Serving模型转换教程，运行下列python文件。
-```
-python tools/inference_to_serving.py --model_dir ch_rec_r34_vd_crnn
-```
-最终会在`serving_client_dir`和`serving_server_dir`生成客户端和服务端的模型配置。其中`serving_server_dir`和`serving_client_dir`的名字可以自定义。最终文件结构如下
-
-```
-/ch_rec_r34_vd_crnn/
-├── serving_client_dir # 客户端配置文件夹
-└── serving_server_dir # 服务端配置文件夹
-```
-
-## 三、文本检测模型Serving推理
-
-启动服务可以根据实际需求选择启动`标准版`或者`快速版`，两种方式的对比如下表：  
-
-|版本|特点|适用场景|
-|-|-|-|
-|标准版|稳定性高，分布式部署|适用于吞吐量大，需要跨机房部署的情况|
-|快速版|部署方便，预测速度快|适用于对预测速度要求高，迭代速度快的场景，Windows用户只能选择快速版|
-
-接下来的命令中，我们会指定快速版和标准版的命令。需要说明的是，标准版只能用Linux平台，快速版可以支持Linux/Windows。
-文本检测模型推理，默认使用DB模型的配置参数，识别默认为CRNN。
-
-配置文件在`params.py`中，我们贴出配置部分，如果需要做改动，也在这个文件内部进行修改。
-
-```
-def read_params():
-    cfg = Config()
-    #use gpu
-    cfg.use_gpu = False # 是否使用GPU
-    cfg.use_pdserving = True # 是否使用paddleserving，必须为True
-
-    #params for text detector
-    cfg.det_algorithm = "DB" # 检测算法， DB/EAST等
-    cfg.det_model_dir = "./det_mv_server/" # 检测算法模型路径
-    cfg.det_max_side_len = 960
-
-    #DB params
-    cfg.det_db_thresh =0.3
-    cfg.det_db_box_thresh =0.5
-    cfg.det_db_unclip_ratio =2.0
-
-    #EAST params
-    cfg.det_east_score_thresh = 0.8
-    cfg.det_east_cover_thresh = 0.1
-    cfg.det_east_nms_thresh = 0.2
-
-    #params for text recognizer
-    cfg.rec_algorithm = "CRNN" # 识别算法， CRNN/RARE等
-    cfg.rec_model_dir = "./ocr_rec_server/" # 识别算法模型路径
-
-    cfg.rec_image_shape = "3, 32, 320"
-    cfg.rec_batch_num = 30
-    cfg.max_text_length = 25
-
-    cfg.rec_char_dict_path = "./ppocr_keys_v1.txt" # 识别算法字典文件
-    cfg.use_space_char = True
-
-    #params for text classifier
-    cfg.use_angle_cls = True # 是否启用分类算法
-    cfg.cls_model_dir = "./ocr_clas_server/" # 分类算法模型路径
-    cfg.cls_image_shape = "3, 48, 192"
-    cfg.label_list = ['0', '180']
-    cfg.cls_batch_num = 30
-    cfg.cls_thresh = 0.9
-
-    return cfg
-```
-与本地预测不同的是，Serving预测需要一个客户端和一个服务端，因此接下来的教程都是两行代码。
-
-在正式执行服务端启动命令之前，先export PYTHONPATH到工程主目录下。
-```
-export PYTHONPATH=$PWD:$PYTHONPATH
-cd deploy/pdserving
-```
-为了方便用户复现Demo程序，我们提供了Chinese and English ultra-lightweight OCR model (8.1M)版本的Serving模型
-```
-wget --no-check-certificate https://paddleocr.bj.bcebos.com/deploy/pdserving/ocr_pdserving_suite.tar.gz
-tar xf ocr_pdserving_suite.tar.gz
-```
-
-### 1. 超轻量中文检测模型推理
-
-超轻量中文检测模型推理，可以执行如下命令启动服务端：
-
-```
-#根据环境只需要启动其中一个就可以
-python det_rpc_server.py #标准版，Linux用户
-python det_local_server.py #快速版，Windows/Linux用户
-```
-
-客户端
-
-```
-python det_web_client.py
-```
-
-
-Serving的推测和本地预测不同点在于，客户端发送请求到服务端，服务端需要检测到文字框之后返回框的坐标，此处没有后处理的图片，只能看到坐标值。
-
-## 四、文本识别模型Serving推理
-
-下面将介绍超轻量中文识别模型推理、基于CTC损失的识别模型推理和基于Attention损失的识别模型推理。对于中文文本识别，建议优先选择基于CTC损失的识别模型，实践中也发现基于Attention损失的效果不如基于CTC损失的识别模型。此外，如果训练时修改了文本的字典，请参考下面的自定义文本识别字典的推理。
-
-### 1. 超轻量中文识别模型推理
-
-超轻量中文识别模型推理，可以执行如下命令启动服务端：
-需要注意params.py中的`--use_gpu`的值
-```
-#根据环境只需要启动其中一个就可以
-python rec_rpc_server.py #标准版，Linux用户
-python rec_local_server.py #快速版，Windows/Linux用户
-```
-如果需要使用CPU版本，还需增加 `--use_gpu False`。
-
-客户端
-
-```
-python rec_web_client.py
-```
-
-![](../imgs_words/ch/word_4.jpg)
-
-执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
-
-```
-{u'result': {u'score': [u'0.89547354'], u'pred_text': ['实力活力']}}
-```
-
-
-
-## 五、方向分类模型推理
-
-下面将介绍方向分类模型推理。
-
-
-
-### 1. 方向分类模型推理
-
-方向分类模型推理， 可以执行如下命令启动服务端：
-需要注意params.py中的`--use_gpu`的值
-```
-#根据环境只需要启动其中一个就可以
-python clas_rpc_server.py #标准版，Linux用户
-python clas_local_server.py #快速版，Windows/Linux用户
-```
-
-客户端
-
-```
-python rec_web_client.py
-```
-
-![](../imgs_words/ch/word_4.jpg)
-
-执行命令后，上面图像的预测结果（分类的方向和得分）会打印到屏幕上，示例如下：
-
-```
-{u'result': {u'direction': [u'0'], u'score': [u'0.9999963']}}
-```
-
-
-## 六、文本检测、方向分类和文字识别串联Serving推理
-
-### 1. 超轻量中文OCR模型推理
-
-在执行预测时，需要通过参数`image_dir`指定单张图像或者图像集合的路径、参数`det_model_dir`,`cls_model_dir`和`rec_model_dir`分别指定检测，方向分类和识别的inference模型路径。参数`use_angle_cls`用于控制是否启用方向分类模型。与本地预测不同的是，为了减少网络传输耗时，可视化识别结果目前不做处理，用户收到的是推理得到的文字字段。
-
-执行如下命令启动服务端：
-需要注意params.py中的`--use_gpu`的值
-```
-#标准版，Linux用户
-#GPU用户
-python -m paddle_serving_server_gpu.serve --model det_infer_server --port 9293 --gpu_id 0
-python -m paddle_serving_server_gpu.serve --model cls_infer_server --port 9294 --gpu_id 0
-python ocr_rpc_server.py
-#CPU用户
-python -m paddle_serving_server.serve --model det_infer_server --port 9293
-python -m paddle_serving_server.serve --model cls_infer_server --port 9294
-python ocr_rpc_server.py
-
-#快速版，Windows/Linux用户
-python ocr_local_server.py
-```
-
-客户端
-
-```
-python rec_web_client.py
-```

doc/doc_ch/update.md

 # 更新
 - 2022.5.9 发布PaddleOCR v2.5。发布内容包括：
-    - [PP-OCRv3](./doc/doc_ch/ppocr_introduction.md#pp-ocrv3)，速度可比情况下，中文场景效果相比于PP-OCRv2再提升5%，英文场景提升11%，80语种多语言模型平均识别准确率提升5%以上；
-    - 半自动标注工具[PPOCRLabelv2](./PPOCRLabel)：新增表格文字图像、图像关键信息抽取任务和不规则文字图像的标注功能；
+    - [PP-OCRv3](./ppocr_introduction.md#pp-ocrv3)，速度可比情况下，中文场景效果相比于PP-OCRv2再提升5%，英文场景提升11%，80语种多语言模型平均识别准确率提升5%以上；
+    - 半自动标注工具[PPOCRLabelv2](../../PPOCRLabel)：新增表格文字图像、图像关键信息抽取任务和不规则文字图像的标注功能；
     - OCR产业落地工具集：打通22种训练部署软硬件环境与方式，覆盖企业90%的训练部署环境需求
-    - 交互式OCR开源电子书[《动手学OCR》](./doc/doc_ch/ocr_book.md)，覆盖OCR全栈技术的前沿理论与代码实践，并配套教学视频。
+    - 交互式OCR开源电子书[《动手学OCR》](./ocr_book.md)，覆盖OCR全栈技术的前沿理论与代码实践，并配套教学视频。
 - 2022.5.7 添加对[Weights & Biases](https://docs.wandb.ai/)训练日志记录工具的支持。
 - 2021.12.21 《OCR十讲》课程开讲，12月21日起每晚八点半线上授课！ 【免费】报名地址：https://aistudio.baidu.com/aistudio/course/introduce/25207
 - 2021.12.21 发布PaddleOCR v2.4。OCR算法新增1种文本检测算法（PSENet），3种文本识别算法（NRTR、SEED、SAR）；文档结构化算法新增1种关键信息提取算法（SDMGR），3种DocVQA算法（LayoutLM、LayoutLMv2，LayoutXLM）。

doc/doc_en/algorithm_det_east_en.md

@@ -40,7 +40,7 @@ Please prepare your environment referring to [prepare the environment](./environ
 
 The above EAST model is trained using the ICDAR2015 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](./dataset/ocr_datasets_en.md).
 
-After the data download is complete, please refer to [Text Detection Training Tutorial](./detection.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+After the data download is complete, please refer to [Text Detection Training Tutorial](./detection_en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
 
 
 <a name="4"></a>

doc/doc_en/algorithm_det_fcenet_en.md

@@ -37,7 +37,7 @@ Please prepare your environment referring to [prepare the environment](./environ
 
 The above FCE model is trained using the CTW1500 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](./dataset/ocr_datasets_en.md).
 
-After the data download is complete, please refer to [Text Detection Training Tutorial](./detection.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+After the data download is complete, please refer to [Text Detection Training Tutorial](./detection_en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
 
 <a name="4"></a>
 ## 4. Inference and Deployment

doc/doc_en/algorithm_det_psenet_en.md

@@ -39,7 +39,7 @@ Please prepare your environment referring to [prepare the environment](./environ
 
 The above PSE model is trained using the ICDAR2015 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](./dataset/ocr_datasets_en.md).
 
-After the data download is complete, please refer to [Text Detection Training Tutorial](./detection.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+After the data download is complete, please refer to [Text Detection Training Tutorial](./detection_en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
 
 <a name="4"></a>
 ## 4. Inference and Deployment

doc/doc_en/algorithm_e2e_pgnet_en.md

@@ -36,7 +36,7 @@ The results of detection and recognition are as follows:
 
 <a name="Environment_Configuration"></a>
 ## 2. Environment Configuration
-Please refer to [Operation Environment Preparation](./environment_en.md) to configure PaddleOCR operating environment first, refer to [PaddleOCR Overview and Project Clone](./paddleOCR_overview_en.md) to clone the project
+Please refer to [Operation Environment Preparation](./environment_en.md) to configure PaddleOCR operating environment first, refer to [Project Clone](./clone_en.md) to clone the project
 
 <a name="Quick_Use"></a>
 ## 3. Quick Use

doc/doc_en/algorithm_overview_en.md

@@ -41,6 +41,12 @@ On Total-Text dataset, the text detection result is as follows:
 | --- | --- | --- | --- | --- | --- |
 |SAST|ResNet50_vd|89.63%|78.44%|83.66%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)|
 
+On CTW1500 dataset, the text detection result is as follows:
+
+|Model|Backbone|Precision|Recall|Hmean| Download link|
+| --- | --- | --- | --- | --- |---|  
+|FCE|ResNet50_dcn|88.39%|82.18%|85.27%| [trained model](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar) |
+
 **Note：** Additional data, like icdar2013, icdar2017, COCO-Text, ArT, was added to the model training of SAST. Download English public dataset in organized format used by PaddleOCR from:
 * [Baidu Drive](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (download code: 2bpi).
 * [Google Drive](https://drive.google.com/drive/folders/1ll2-XEVyCQLpJjawLDiRlvo_i4BqHCJe?usp=sharing)

doc/doc_en/algorithm_rec_aster_en.md

@@ -33,13 +33,13 @@ Using MJSynth and SynthText two text recognition datasets for training, and eval
 
 <a name="2"></a>
 ## 2. Environment
-Please refer to ["Environment Preparation"](./environment.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](./clone.md) to clone the project code.
+Please refer to ["Environment Preparation"](./environment_en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](./clone_en.md) to clone the project code.
 
 
 <a name="3"></a>
 ## 3. Model Training / Evaluation / Prediction
 
-Please refer to [Text Recognition Tutorial](./recognition.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+Please refer to [Text Recognition Tutorial](./recognition_en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
 
 Training: