Merge branch 'dygraph' into android_demo

3ff94920 · zhoujun · GitHub · 6a160e97 · f2fc1a3f · 3ff94920
Unverified Commit 3ff94920 authored Apr 09, 2021 by zhoujun Committed by GitHub Apr 09, 2021
20 changed files
--- a/deploy/pdserving/web_service.py
+++ b/deploy/pdserving/web_service.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+try:
+    from paddle_serving_server_gpu.web_service import WebService, Op
+except ImportError:
+    from paddle_serving_server.web_service import WebService, Op
+import logging
+import numpy as np
+import cv2
+import base64
+# from paddle_serving_app.reader import OCRReader
+from ocr_reader import OCRReader, DetResizeForTest
+from paddle_serving_app.reader import Sequential, ResizeByFactor
+from paddle_serving_app.reader import Div, Normalize, Transpose
+from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
+_LOGGER = logging.getLogger()
+class DetOp(Op):
+    def init_op(self):
+        self.det_preprocess = Sequential([
+            DetResizeForTest(), Div(255),
+            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
+                (2, 0, 1))
+        ])
+        self.filter_func = FilterBoxes(10, 10)
+        self.post_func = DBPostProcess({
+            "thresh": 0.3,
+            "box_thresh": 0.5,
+            "max_candidates": 1000,
+            "unclip_ratio": 1.5,
+            "min_size": 3
+        })
+    def preprocess(self, input_dicts, data_id, log_id):
+        (_, input_dict), = input_dicts.items()
+        data = base64.b64decode(input_dict["image"].encode('utf8'))
+        data = np.fromstring(data, np.uint8)
+        # Note: class variables(self.var) can only be used in process op mode
+        im = cv2.imdecode(data, cv2.IMREAD_COLOR)
+        self.im = im
+        self.ori_h, self.ori_w, _ = im.shape
+        det_img = self.det_preprocess(self.im)
+        _, self.new_h, self.new_w = det_img.shape
+        print("det image shape", det_img.shape)
+        return {"x": det_img[np.newaxis, :].copy()}, False, None, ""
+    def postprocess(self, input_dicts, fetch_dict, log_id):
+        print("input_dicts: ", input_dicts)
+        det_out = fetch_dict["save_infer_model/scale_0.tmp_1"]
+        ratio_list = [
+            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
+        ]
+        dt_boxes_list = self.post_func(det_out, [ratio_list])
+        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
+        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
+        print("out dict", out_dict["dt_boxes"])
+        return out_dict, None, ""
+class RecOp(Op):
+    def init_op(self):
+        self.ocr_reader = OCRReader(
+            char_dict_path="../../ppocr/utils/ppocr_keys_v1.txt")
+        self.get_rotate_crop_image = GetRotateCropImage()
+        self.sorted_boxes = SortedBoxes()
+    def preprocess(self, input_dicts, data_id, log_id):
+        (_, input_dict), = input_dicts.items()
+        im = input_dict["image"]
+        dt_boxes = input_dict["dt_boxes"]
+        dt_boxes = self.sorted_boxes(dt_boxes)
+        feed_list = []
+        img_list = []
+        max_wh_ratio = 0
+        for i, dtbox in enumerate(dt_boxes):
+            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
+            img_list.append(boximg)
+            h, w = boximg.shape[0:2]
+            wh_ratio = w * 1.0 / h
+            max_wh_ratio = max(max_wh_ratio, wh_ratio)
+        _, w, h = self.ocr_reader.resize_norm_img(img_list[0],
+                                                  max_wh_ratio).shape
+        imgs = np.zeros((len(img_list), 3, w, h)).astype('float32')
+        for id, img in enumerate(img_list):
+            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
+            imgs[id] = norm_img
+        print("rec image shape", imgs.shape)
+        feed = {"x": imgs.copy()}
+        return feed, False, None, ""
+    def postprocess(self, input_dicts, fetch_dict, log_id):
+        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
+        res_lst = []
+        for res in rec_res:
+            res_lst.append(res[0])
+        res = {"res": str(res_lst)}
+        return res, None, ""
+class OcrService(WebService):
+    def get_pipeline_response(self, read_op):
+        det_op = DetOp(name="det", input_ops=[read_op])
+        rec_op = RecOp(name="rec", input_ops=[det_op])
+        return rec_op
+uci_service = OcrService(name="ocr")
+uci_service.prepare_pipeline_config("config.yml")
+uci_service.run_service()
--- a/deploy/slim/prune/README.md
+++ b/deploy/slim/prune/README.md
+## 介绍
+复杂的模型有利于提高模型的性能，但也导致模型中存在一定冗余，模型裁剪通过移出网络模型中的子模型来减少这种冗余，达到减少模型计算复杂度，提高模型推理性能的目的。
+本教程将介绍如何使用飞桨模型压缩库PaddleSlim做PaddleOCR模型的压缩。
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)集成了模型剪枝、量化（包括量化训练和离线量化）、蒸馏和神经网络搜索等多种业界常用且领先的模型压缩功能，如果您感兴趣，可以关注并了解。
+在开始本教程之前，建议先了解：
+1. [PaddleOCR模型的训练方法](../../../doc/doc_ch/quickstart.md)
+2. [模型裁剪教程](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+## 快速开始
+模型裁剪主要包括四个步骤：
+1. 安装 PaddleSlim
+2. 准备训练好的模型
+3. 敏感度分析、裁剪训练
+4. 导出模型、预测部署
+### 1. 安装PaddleSlim
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+git checkout develop
+cd Paddleslim
+python3 setup.py install
+```
+### 2. 获取预训练模型
+模型裁剪需要加载事先训练好的模型，PaddleOCR也提供了一系列(模型)[../../../doc/doc_ch/models_list.md]，开发者可根据需要自行选择模型或使用自己的模型。
+### 3. 敏感度分析训练
+加载预训练模型后，通过对现有模型的每个网络层进行敏感度分析，得到敏感度文件：sen.pickle，可以通过PaddleSlim提供的[接口](https://github.com/PaddlePaddle/PaddleSlim/blob/9b01b195f0c4bc34a1ab434751cb260e13d64d9e/paddleslim/dygraph/prune/filter_pruner.py#L75)加载文件，获得各网络层在不同裁剪比例下的精度损失。从而了解各网络层冗余度，决定每个网络层的裁剪比例。
+敏感度文件内容格式：
+    sen.pickle(Dict){
+            'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+            'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+        }
+    例子：
+        {
+            'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+            'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+        }
+加载敏感度文件后会返回一个字典，字典中的keys为网络模型参数模型的名字，values为一个字典，里面保存了相应网络层的裁剪敏感度信息。例如在例子中，conv10_expand_weights所对应的网络层在裁掉10%的卷积核后模型性能相较原模型会下降0.65%，详细信息可见[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+进入PaddleOCR根目录，通过以下命令对模型进行敏感度分析训练：
+```bash
+python3.7 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights="your trained model"
+```
+### 4. 导出模型、预测部署
+在得到裁剪训练保存的模型后，我们可以将其导出为inference_model：
+```bash
+pytho3.7 deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights=./output/det_db/best_accuracy  Global.save_inference_dir=inference_model
+```
+inference model的预测和部署参考：
+1. [inference model python端预测](../../../doc/doc_ch/inference.md)
+2. [inference model C++预测](../../cpp_infer/readme.md)
--- a/deploy/slim/prune/README_en.md
+++ b/deploy/slim/prune/README_en.md
+## Introduction
+Generally, a more complex model would achive better performance in the task, but it also leads to some redundancy in the model. Model Pruning is a technique that reduces this redundancy by removing the sub-models in the neural network model, so as to reduce model calculation complexity and improve model inference performance.
+This example uses PaddleSlim provided[APIs of Pruning](https://paddlepaddle.github.io/PaddleSlim/api/prune_api/) to compress the OCR model.
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim), an open source library which integrates model pruning, quantization (including quantization training and offline quantization), distillation, neural network architecture search, and many other commonly used and leading model compression technique in the industry.
+It is recommended that you could understand following pages before reading this example：
+1. [PaddleOCR training methods](../../../doc/doc_ch/quickstart.md)
+2. [The demo of prune](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+## Quick start
+Five steps for OCR model prune:
+1. Install PaddleSlim
+2. Prepare the trained model
+3. Sensitivity analysis and tailoring training
+4. Export model, predict deployment
+### 1. Install PaddleSlim
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+git checkout develop
+cd Paddleslim
+python3 setup.py install
+```
+### 2. Download Pretrain Model
+Model prune needs to load pre-trained models.
+PaddleOCR also provides a series of (models)[../../../doc/doc_en/models_list_en.md]. Developers can choose their own models or use their own models according to their needs.
+### 3. Pruning sensitivity analysis
+  After the pre-training model is loaded, sensitivity analysis is performed on each network layer of the model to understand the redundancy of each network layer, and save a sensitivity file which named: sen.pickle.  After that, user could load the sensitivity file via the [methods provided by PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/prune/sensitive.py#L221) and determining the pruning ratio of each network layer automatically. For specific details of sensitivity analysis, see：[Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/tutorials/image_classification_sensitivity_analysis_tutorial.md)
+  The data format of sensitivity file：
+      sen.pickle(Dict){
+              'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+              'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+          }
+      example：
+          {
+              'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+              'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+          }
+  The function would return a dict after loading the sensitivity file. The keys of the dict are name of parameters in each layer. And the value of key is the information about pruning sensitivity of correspoding layer. In example, pruning 10% filter of the layer corresponding to conv10_expand_weights would lead to 0.65% degradation of model performance. The details could be seen at: [Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+Enter the PaddleOCR root directory，perform sensitivity analysis on the model with the following command：
+```bash
+python3.7 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrain_weights="your trained model"
+```
+### 5.  Export inference model and deploy it
+We can export the pruned model as inference_model for deployment:
+```bash
+python deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml  -o Global.pretrain_weights=./output/det_db/best_accuracy Global.test_batch_size_per_card=1 Global.save_inference_dir=inference_model
+```
+Reference for prediction and deployment of inference model:
+1. [inference model python prediction](../../../doc/doc_en/inference_en.md)
+2. [inference model C++ prediction](../../cpp_infer/readme_en.md)
--- a/deploy/slim/prune/export_prune_model.py
+++ b/deploy/slim/prune/export_prune_model.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+import os
+import sys
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, '..', '..', '..'))
+sys.path.append(os.path.join(__dir__, '..', '..', '..', 'tools'))
+import paddle
+from ppocr.data import build_dataloader
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import init_model
+import tools.program as program
+def main(config, device, logger, vdl_writer):
+    global_config = config['Global']
+    # build dataloader
+    valid_dataloader = build_dataloader(config, 'Eval', device, logger)
+    # build post process
+    post_process_class = build_post_process(config['PostProcess'],
+                                            global_config)
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, 'character'):
+        char_num = len(getattr(post_process_class, 'character'))
+        config['Architecture']["Head"]['out_channels'] = char_num
+    model = build_model(config['Architecture'])
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs before pruning: {flops}")
+    from paddleslim.dygraph import FPGMFilterPruner
+    model.train()
+    pruner = FPGMFilterPruner(model, [1, 3, 640, 640])
+    # build metric
+    eval_class = build_metric(config['Metric'])
+    def eval_fn():
+        metric = program.eval(model, valid_dataloader, post_process_class,
+                              eval_class)
+        logger.info(f"metric['hmean']: {metric['hmean']}")
+        return metric['hmean']
+    params_sensitive = pruner.sensitive(
+        eval_func=eval_fn,
+        sen_file="./sen.pickle",
+        skip_vars=[
+            "conv2d_57.w_0", "conv2d_transpose_2.w_0", "conv2d_transpose_3.w_0"
+        ])
+    logger.info(
+        "The sensitivity analysis results of model parameters saved in sen.pickle"
+    )
+    # calculate pruned params's ratio
+    params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+    for key in params_sensitive.keys():
+        logger.info(f"{key}, {params_sensitive[key]}")
+    plan = pruner.prune_vars(params_sensitive, [0])
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs after pruning: {flops}")
+    # load pretrain model
+    pre_best_model_dict = init_model(config, model, logger, None)
+    metric = program.eval(model, valid_dataloader, post_process_class,
+                          eval_class)
+    logger.info(f"metric['hmean']: {metric['hmean']}")
+    # start export model
+    from paddle.jit import to_static
+    infer_shape = [3, -1, -1]
+    if config['Architecture']['model_type'] == "rec":
+        infer_shape = [3, 32, -1]  # for rec model, H must be 32
+        if 'Transform' in config['Architecture'] and config['Architecture'][
+                'Transform'] is not None and config['Architecture'][
+                    'Transform']['name'] == 'TPS':
+            logger.info(
+                'When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training'
+            )
+            infer_shape[-1] = 100
+    model = to_static(
+        model,
+        input_spec=[
+            paddle.static.InputSpec(
+                shape=[None] + infer_shape, dtype='float32')
+        ])
+    save_path = '{}/inference'.format(config['Global']['save_inference_dir'])
+    paddle.jit.save(model, save_path)
+    logger.info('inference model is saved to {}'.format(save_path))
+if __name__ == '__main__':
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)
--- a/deploy/slim/prune/sensitivity_anal.py
+++ b/deploy/slim/prune/sensitivity_anal.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+import os
+import sys
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, '..', '..', '..'))
+sys.path.append(os.path.join(__dir__, '..', '..', '..', 'tools'))
+import paddle
+import paddle.distributed as dist
+from ppocr.data import build_dataloader
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import init_model
+import tools.program as program
+dist.get_world_size()
+def get_pruned_params(parameters):
+    params = []
+    for param in parameters:
+        if len(
+                param.shape
+        ) == 4 and 'depthwise' not in param.name and 'transpose' not in param.name and "conv2d_57" not in param.name and "conv2d_56" not in param.name:
+            params.append(param.name)
+    return params
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config['Global']['distributed']:
+        dist.init_parallel_env()
+    global_config = config['Global']
+    # build dataloader
+    train_dataloader = build_dataloader(config, 'Train', device, logger)
+    if config['Eval']:
+        valid_dataloader = build_dataloader(config, 'Eval', device, logger)
+    else:
+        valid_dataloader = None
+    # build post process
+    post_process_class = build_post_process(config['PostProcess'],
+                                            global_config)
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, 'character'):
+        char_num = len(getattr(post_process_class, 'character'))
+        config['Architecture']["Head"]['out_channels'] = char_num
+    model = build_model(config['Architecture'])
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs before pruning: {flops}")
+    from paddleslim.dygraph import FPGMFilterPruner
+    model.train()
+    pruner = FPGMFilterPruner(model, [1, 3, 640, 640])
+    # build loss
+    loss_class = build_loss(config['Loss'])
+    # build optim
+    optimizer, lr_scheduler = build_optimizer(
+        config['Optimizer'],
+        epochs=config['Global']['epoch_num'],
+        step_each_epoch=len(train_dataloader),
+        parameters=model.parameters())
+    # build metric
+    eval_class = build_metric(config['Metric'])
+    # load pretrain model
+    pre_best_model_dict = init_model(config, model, logger, optimizer)
+    logger.info('train dataloader has {} iters, valid dataloader has {} iters'.
+                format(len(train_dataloader), len(valid_dataloader)))
+    # build metric
+    eval_class = build_metric(config['Metric'])
+    logger.info('train dataloader has {} iters, valid dataloader has {} iters'.
+                format(len(train_dataloader), len(valid_dataloader)))
+    def eval_fn():
+        metric = program.eval(model, valid_dataloader, post_process_class,
+                              eval_class)
+        logger.info(f"metric['hmean']: {metric['hmean']}")
+        return metric['hmean']
+    params_sensitive = pruner.sensitive(
+        eval_func=eval_fn,
+        sen_file="./sen.pickle",
+        skip_vars=[
+            "conv2d_57.w_0", "conv2d_transpose_2.w_0", "conv2d_transpose_3.w_0"
+        ])
+    logger.info(
+        "The sensitivity analysis results of model parameters saved in sen.pickle"
+    )
+    # calculate pruned params's ratio
+    params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+    for key in params_sensitive.keys():
+        logger.info(f"{key}, {params_sensitive[key]}")
+    plan = pruner.prune_vars(params_sensitive, [0])
+    for param in model.parameters():
+        if ("weights" in param.name and "conv" in param.name) or (
+                "w_0" in param.name and "conv2d" in param.name):
+            logger.info(f"{param.name}: {param.shape}")
+    flops = paddle.flops(model, [1, 3, 640, 640])
+    logger.info(f"FLOPs after pruning: {flops}")
+    # start train
+    program.train(config, train_dataloader, valid_dataloader, device, model,
+                  loss_class, optimizer, lr_scheduler, post_process_class,
+                  eval_class, pre_best_model_dict, logger, vdl_writer)
+if __name__ == '__main__':
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -28,7 +28,9 @@ PaddleOCR开源的文本检测算法列表：
 | --- | --- | --- | --- | --- | --- |
 |SAST|ResNet50_vd|89.63%|78.44%|83.66%|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)|
-**说明：** SAST模型训练额外加入了icdar2013、icdar2017、COCO-Text、ArT等公开数据集进行调优。PaddleOCR用到的经过整理格式的英文公开数据集下载：[百度云地址](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (提取码: 2bpi)
+**说明：** SAST模型训练额外加入了icdar2013、icdar2017、COCO-Text、ArT等公开数据集进行调优。PaddleOCR用到的经过整理格式的英文公开数据集下载：
+* [百度云地址](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (提取码: 2bpi)
+* [Google Drive下载地址](https://drive.google.com/drive/folders/1ll2-XEVyCQLpJjawLDiRlvo_i4BqHCJe?usp=sharing)
 PaddleOCR文本检测算法的训练和使用请参考文档教程中[模型训练/评估中的文本检测部分](./detection.md)。

--- a/doc/doc_ch/angle_class.md
+++ b/doc/doc_ch/angle_class.md
 ## 文字角度分类
+### 方法介绍
+文字角度分类主要用于图片非0度的场景下，在这种场景下需要对图片里检测到的文本行进行一个转正的操作。在PaddleOCR系统内，
+文字检测之后得到的文本行图片经过仿射变换之后送入识别模型，此时只需要对文字进行一个0和180度的角度分类，因此PaddleOCR内置的
+文字角度分类器**只支持了0和180度的分类**。如果想支持更多角度，可以自己修改算法进行支持。
+0和180度数据样本例子：
+![](../imgs_results/angle_class_example.jpg)
 ### 数据准备
@@ -13,7 +21,7 @@ ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/cls/dataset
 请参考下文组织您的数据。
 - 训练集
-首先请将训练图片放入同一个文件夹（train_images），并用一个txt文件（cls_gt_train.txt）记录图片路径和标签。
+首先建议将训练图片放入同一个文件夹，并用一个txt文件（cls_gt_train.txt）记录图片路径和标签。
 **注意：** 默认请将图片路径和图片标签用 `\t` 分割，如用其他方式分割将造成训练报错
@@ -21,8 +29,8 @@ ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/cls/dataset
 ```
 " 图像文件名                 图像标注信息 "
-train/word_001.jpg   0
+train/cls/train/word_001.jpg   0
-train/word_002.jpg   180
+train/cls/train/word_002.jpg   180
 ```
 最终训练集应有如下文件结构：

--- a/doc/doc_ch/inference.md
+++ b/doc/doc_ch/inference.md
@@ -2,16 +2,18 @@
 # 基于Python预测引擎推理
 inference 模型（`paddle.jit.save`保存的模型）
-一般是模型训练完成后保存的固化模型，多用于预测部署。训练过程中保存的模型是checkpoints模型，保存的是模型的参数，多用于恢复训练等。
+一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。
-与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合与实际系统集成。
+训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。
+与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
-接下来首先介绍如何将训练的模型转换成inference模型，然后将依次介绍文本检测、文本角度分类器、文本识别以及三者串联基于预测引擎推理。
+接下来首先介绍如何将训练的模型转换成inference模型，然后将依次介绍文本检测、文本角度分类器、文本识别以及三者串联在CPU、GPU上的预测方法。
 - [一、训练模型转inference模型](#训练模型转inference模型)
    - [检测模型转inference模型](#检测模型转inference模型)
    - [识别模型转inference模型](#识别模型转inference模型)  
-    - [方向分类模型转inference模型](#方向分类模型转inference模型)  
+    - [方向分类模型转inference模型](#方向分类模型转inference模型)
+    - [端到端模型转inference模型](#端到端模型转inference模型)
 - [二、文本检测模型推理](#文本检测模型推理)
    - [1. 超轻量中文检测模型推理](#超轻量中文检测模型推理)
@@ -26,10 +28,13 @@ inference 模型（`paddle.jit.save`保存的模型）
    - [4. 自定义文本识别字典的推理](#自定义文本识别字典的推理)
    - [5. 多语言模型的推理](#多语言模型的推理)
- [四、方向分类模型推理](#方向识别模型推理)
+- [四、端到端模型推理](#端到端模型推理)
+    - [1. PGNet端到端模型推理](#PGNet端到端模型推理)
+- [五、方向分类模型推理](#方向识别模型推理)
    - [1. 方向分类模型推理](#方向分类模型推理)
- [五、文本检测、方向分类和文字识别串联推理](#文本检测、方向分类和文字识别串联推理)
+- [六、文本检测、方向分类和文字识别串联推理](#文本检测、方向分类和文字识别串联推理)
    - [1. 超轻量中文OCR模型推理](#超轻量中文OCR模型推理)
    - [2. 其他模型推理](#其他模型推理)
@@ -117,6 +122,32 @@ python3 tools/export_model.py -c configs/cls/cls_mv3.yml -o Global.pretrained_mo
    ├── inference.pdiparams.info    # 分类inference模型的参数信息，可忽略
    └── inference.pdmodel           # 分类inference模型的program文件
 ```
+<a name="端到端模型转inference模型"></a>
+### 端到端模型转inference模型
+下载端到端模型：
+```
+wget -P ./ch_lite/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar && tar xf ./ch_lite/ch_ppocr_mobile_v2.0_cls_train.tar -C ./ch_lite/
+```
+端到端模型转inference模型与检测的方式相同，如下：
+```
+# -c 后面设置训练算法的yml配置文件
+# -o 配置可选参数
+# Global.pretrained_model 参数设置待转换的训练模型地址，不用添加文件后缀 .pdmodel，.pdopt或.pdparams。
+# Global.load_static_weights 参数需要设置为 False。
+# Global.save_inference_dir参数设置转换的模型将保存的地址。
+python3 tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./ch_lite/ch_ppocr_mobile_v2.0_cls_train/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./inference/e2e/
+```
+转换成功后，在目录下有三个文件：
+```
+/inference/e2e/
+    ├── inference.pdiparams         # 分类inference模型的参数文件
+    ├── inference.pdiparams.info    # 分类inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 分类inference模型的program文件
+```
 <a name="文本检测模型推理"></a>
 ## 二、文本检测模型推理
@@ -140,7 +171,7 @@ python3 tools/infer/predict_det.py --image_dir="./doc/imgs/00018069.jpg" --det_m
 ![](../imgs_results/det_res_00018069.jpg)
 通过参数`limit_type`和`det_limit_side_len`来对图片的尺寸进行限制，
-`litmit_type`可选参数为[`max`, `min`]，
+`limit_type`可选参数为[`max`, `min`]，
 `det_limit_size_len` 为正整数，一般设置为32 的倍数，比如960。
 参数默认设置为`limit_type='max', det_limit_side_len=960`。表示网络输入图像的最长边不能超过960，
@@ -331,8 +362,38 @@ python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/korean/1.jpg" -
 Predicts of ./doc/imgs_words/korean/1.jpg:('바탕으로', 0.9948904)
 ```
+<a name="端到端模型推理"></a>
+## 四、端到端模型推理
+端到端模型推理，默认使用PGNet模型的配置参数。当不使用PGNet模型时，在推理时，需要通过传入相应的参数进行算法适配，细节参考下文。
+<a name="PGNet端到端模型推理"></a>
+### 1. PGNet端到端模型推理
+#### (1). 四边形文本检测模型（ICDAR2015）  
+首先将PGNet端到端训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例([模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar))，可以使用如下命令进行转换：
+```
+python3 tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./en_server_pgnetA/iter_epoch_450 Global.load_static_weights=False Global.save_inference_dir=./inference/e2e
+```
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"`**，可以执行如下命令：
+```
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img_10.jpg" --e2e_model_dir="./inference/e2e/"  --e2e_pgnet_polygon=False
+```
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+![](../imgs_results/e2e_res_img_10_pgnet.jpg)
+#### (2). 弯曲文本检测模型（Total-Text）  
+和四边形文本检测模型共用一个推理模型
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"`，同时，还需要增加参数`--e2e_pgnet_polygon=True`，**可以执行如下命令：
+```
+python3.7 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_polygon=True
+```
+可视化文本端到端结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+![](../imgs_results/e2e_res_img623_pgnet.jpg)
 <a name="方向分类模型推理"></a>
-## 四、方向分类模型推理
+## 五、方向分类模型推理
 下面将介绍方向分类模型推理。
@@ -357,7 +418,7 @@ Predicts of ./doc/imgs_words/ch/word_4.jpg:['0', 0.9999982]
 ```
 <a name="文本检测、方向分类和文字识别串联推理"></a>
-## 五、文本检测、方向分类和文字识别串联推理
+## 六、文本检测、方向分类和文字识别串联推理
 <a name="超轻量中文OCR模型推理"></a>
 ### 1. 超轻量中文OCR模型推理

--- a/doc/doc_ch/installation.md
+++ b/doc/doc_ch/installation.md
@@ -30,7 +30,7 @@ sudo nvidia-docker run --name ppocr -v $PWD:/paddle --shm-size=64G --network=hos
 sudo docker container exec -it ppocr /bin/bash
 ```
-**2. 安装PaddlePaddle Fluid v2.0**
+**2. 安装PaddlePaddle 2.0**
 ```
 pip3 install --upgrade pip

--- a/doc/doc_ch/pgnet.md
+++ b/doc/doc_ch/pgnet.md
+# 端对端OCR算法-PGNet
+- [一、简介](#简介)
+- [二、环境配置](#环境配置)
+- [三、快速使用](#快速使用)
+- [四、快速训练](#开始训练)
+- [五、预测推理](#预测推理)
+<a name="简介"></a>
+##简介
+OCR算法可以分为两阶段算法和端对端的算法。二阶段OCR算法一般分为两个部分，文本检测和文本识别算法，文件检测算法从图像中得到文本行的检测框，然后识别算法去识别文本框中的内容。而端对端OCR算法可以在一个算法中完成文字检测和文字识别，其基本思想是设计一个同时具有检测单元和识别模块的模型，共享其中两者的CNN特征，并联合训练。由于一个算法即可完成文字识别，端对端模型更小，速度更快。
+### PGNet算法介绍
+近些年来，端对端OCR算法得到了良好的发展，包括MaskTextSpotter系列、TextSnake、TextDragon、PGNet系列等算法。在这些算法中，PGNet算法具备其他算法不具备的优势，包括：
+- 设计PGNet loss指导训练，不需要字符级别的标注
+- 不需要NMS和ROI相关操作，加速预测
+- 提出预测文本行内的阅读顺序模块；
+- 提出基于图的修正模块（GRM）来进一步提高模型识别性能
+- 精度更高，预测速度更快
+PGNet算法细节详见[论文](https://www.aaai.org/AAAI21Papers/AAAI-2885.WangP.pdf)， 算法原理图如下所示：
+![](../pgnet_framework.png)
+输入图像经过特征提取送入四个分支，分别是：文本边缘偏移量预测TBO模块，文本中心线预测TCL模块，文本方向偏移量预测TDO模块，以及文本字符分类图预测TCC模块。
+其中TBO以及TCL的输出经过后处理后可以得到文本的检测结果，TCL、TDO、TCC负责文本识别。
+其检测识别效果图如下：
+![](../imgs_results/e2e_res_img293_pgnet.png)
+![](../imgs_results/e2e_res_img295_pgnet.png)
+<a name="环境配置"></a>
+##环境配置
+请先参考[快速安装](./installation.md)配置PaddleOCR运行环境。
+*注意：也可以通过 whl 包安装使用PaddleOCR，具体参考[Paddleocr Package使用说明](./whl.md)。*
+<a name="快速使用"></a>
+##快速使用
+### inference模型下载
+本节以训练好的端到端模型为例，快速使用模型预测，首先下载训练好的端到端inference模型[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar)
+```
+mkdir inference && cd inference
+# 下载英文端到端模型并解压
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar && tar xf e2e_server_pgnetA_infer.tar
+```
+* windows 环境下如果没有安装wget,下载模型时可将链接复制到浏览器中下载，并解压放置在相应目录下
+解压完毕后应有如下文件结构：
+```
+├── e2e_server_pgnetA_infer
+│   ├── inference.pdiparams
+│   ├── inference.pdiparams.info
+│   └── inference.pdmodel
+```
+### 单张图像或者图像集合预测
+```bash
+# 预测image_dir指定的单张图像
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_polygon=True
+# 预测image_dir指定的图像集合
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_polygon=True
+# 如果想使用CPU进行预测，需设置use_gpu参数为False
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_polygon=True --use_gpu=False
+```
+<a name="开始训练"></a>
+##开始训练
+本节以totaltext数据集为例，介绍PaddleOCR中端到端模型的训练、评估与测试。
+###数据形式为icdar, 十六点标注数据
+解压数据集和下载标注文件后，PaddleOCR/train_data/total_text/train/ 有两个文件夹，分别是：
+```
+/PaddleOCR/train_data/total_text/train/
+  |- rgb/            total_text数据集的训练数据
+      |- gt_0.png
+      | ...  
+  |- total_text.txt  total_text数据集的训练标注
+```
+提供的标注文件格式如下，中间用"\t"分隔：
+```
+" 图像文件名                    json.dumps编码的图像标注信息"
+rgb/gt_0.png    [{"transcription": "EST", "points": [[1004.0,689.0],[1019.0,698.0],[1034.0,708.0],[1049.0,718.0],[1064.0,728.0],[1079.0,738.0],[1095.0,748.0],[1094.0,774.0],[1079.0,765.0],[1065.0,756.0],[1050.0,747.0],[1036.0,738.0],[1021.0,729.0],[1007.0,721.0]]}, {...}]
+```
+json.dumps编码前的图像标注信息是包含多个字典的list，字典中的 `points` 表示文本框的四个点的坐标(x, y)，从左上角的点开始顺时针排列。
+`transcription` 表示当前文本框的文字，**当其内容为“###”时，表示该文本框无效，在训练时会跳过。**
+如果您想在其他数据集上训练，可以按照上述形式构建标注文件。
+### 快速启动训练
+模型训练一般分两步骤进行，第一步可以选择用合成数据训练，第二步加载第一步训练好的模型训练，这边我们提供了第一步训练好的模型，可以直接加载，从第二步开始训练
+[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/train_step1.tar)
+```shell
+cd PaddleOCR/
+下载ResNet50_vd的动态图预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/train_step1.tar
+可以得到以下的文件格式
+./pretrain_models/train_step1/
+  └─ best_accuracy.pdopt
+  └─ best_accuracy.states
+  └─ best_accuracy.pdparams
+```
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+```shell
+# 单机单卡训练 e2e 模型
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy Global.load_static_weights=False
+# 单机多卡训练，通过 --gpus 参数设置使用的GPU ID
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy  Global.load_static_weights=False
+```
+上述指令中，通过-c 选择训练使用configs/e2e/e2e_r50_vd_pg.yml配置文件。
+有关配置文件的详细解释，请参考[链接](./config.md)。
+您也可以通过-o参数在不需要修改yml文件的情况下，改变训练的参数，比如，调整训练的学习率为0.0001
+```shell
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Optimizer.base_lr=0.0001
+```
+#### 断点训练
+如果训练程序中断，如果希望加载训练中断的模型从而恢复训练，可以通过指定Global.checkpoints指定要加载的模型路径：
+```shell
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.checkpoints=./your/trained/model
+```
+**注意**：`Global.checkpoints`的优先级高于`Global.pretrain_weights`的优先级，即同时指定两个参数时，优先加载`Global.checkpoints`指定的模型，如果`Global.checkpoints`指定的模型路径有误，会加载`Global.pretrain_weights`指定的模型。
+<a name="预测推理"></a>
+## 预测推理
+PaddleOCR计算三个OCR端到端相关的指标，分别是：Precision、Recall、Hmean。
+运行如下代码，根据配置文件`e2e_r50_vd_pg.yml`中`save_res_path`指定的测试集检测结果文件，计算评估指标。
+评估时设置后处理参数`max_side_len=768`，使用不同数据集、不同模型训练，可调整参数进行优化
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Global.checkpoints`指向保存的参数文件。
+```shell
+python3 tools/eval.py -c configs/e2e/e2e_r50_vd_pg.yml  -o Global.checkpoints="{path/to/weights}/best_accuracy"
+```
+### 测试端到端效果
+测试单张图像的端到端识别效果
+```shell
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false
+```
+测试文件夹下所有图像的端到端识别效果
+```shell
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/det_db/best_accuracy" Global.load_static_weights=false
+```
+###转为推理模型
+### (1). 四边形文本检测模型（ICDAR2015）  
+首先将PGNet端到端训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，以英文数据集训练的模型为例[模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar) ，可以使用如下命令进行转换：
+```
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar && tar xf en_server_pgnetA.tar
+python3 tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./en_server_pgnetA/iter_epoch_450 Global.load_static_weights=False Global.save_inference_dir=./inference/e2e
+```
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"`**，可以执行如下命令：
+```
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img_10.jpg" --e2e_model_dir="./inference/e2e/"  --e2e_pgnet_polygon=False
+```
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+![](../imgs_results/e2e_res_img_10_pgnet.jpg)
+### (2). 弯曲文本检测模型（Total-Text）
+对于弯曲文本样例
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"`，同时，还需要增加参数`--e2e_pgnet_polygon=True`，**可以执行如下命令：
+```
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_polygon=True
+```
+可视化文本端到端结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+![](../imgs_results/e2e_res_img623_pgnet.jpg)
--- a/doc/doc_ch/recognition.md
+++ b/doc/doc_ch/recognition.md
 ## 文字识别
- [一、数据准备](#数据准备)
+- [1 数据准备](#数据准备)
-    - [数据下载](#数据下载)
+    - [1.1 自定义数据集](#自定义数据集)
-    - [自定义数据集](#自定义数据集)  
+    - [1.2 数据下载](#数据下载)
-    - [字典](#字典)  
+    - [1.3 字典](#字典)  
-    - [支持空格](#支持空格)
+    - [1.4 支持空格](#支持空格)
- [二、启动训练](#启动训练)
+- [2 启动训练](#启动训练)
-    - [1. 数据增强](#数据增强)
+    - [2.1 数据增强](#数据增强)
-    - [2. 训练](#训练)
+    - [2.2 训练](#训练)
-    - [3. 小语种](#小语种)
+    - [2.3 小语种](#小语种)
- [三、评估](#评估)
+- [3 评估](#评估)
- [四、预测](#预测)
+- [4 预测](#预测)
-    - [1. 训练引擎预测](#训练引擎预测)
+    - [4.1 训练引擎预测](#训练引擎预测)
 <a name="数据准备"></a>
-### 数据准备
+### 1. 数据准备
-PaddleOCR 支持两种数据格式: `lmdb` 用于训练公开数据，调试算法; `通用数据` 训练自己的数据:
+PaddleOCR 支持两种数据格式:
+ - `lmdb` 用于训练以lmdb格式存储的数据集;
-请按如下步骤设置数据集：
+ - `通用数据` 用于训练以文本文件存储的数据集:
 训练数据的默认存储路径是 `PaddleOCR/train_data`,如果您的磁盘上已有数据集，只需创建软链接至数据集目录：
 ```
+# linux and mac os
 ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
 ```
-<a name="数据下载"></a>
+<a name="准备数据集"></a>
-* 数据下载
+#### 1.1 自定义数据集
+下面以通用数据集为例， 介绍如何准备数据集：
-若您本地没有数据集，可以在官网下载 [icdar2015](http://rrc.cvc.uab.es/?ch=4&com=downloads) 数据，用于快速验证。也可以参考[DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here)，下载 benchmark 所需的lmdb格式数据集。
-如果希望复现SRN的论文指标，需要下载离线[增广数据](https://pan.baidu.com/s/1-HSZ-ZVdqBF2HaBZ5pRAKA),提取码: y3ry。增广数据是由MJSynth和SynthText做旋转和扰动得到的。数据下载完成后请解压到 {your_path}/PaddleOCR/train_data/data_lmdb_release/training/ 路径下。
-<a name="自定义数据集"></a>
+* 训练集
-* 使用自己数据集
-若您希望使用自己的数据进行训练，请参考下文组织您的数据。
+建议将训练图片放入同一个文件夹，并用一个txt文件（rec_gt_train.txt）记录图片路径和标签，txt文件里的内容如下:
- 训练集
+**注意：** txt文件中默认请将图片路径和图片标签用 \t 分割，如用其他方式分割将造成训练报错。
-首先请将训练图片放入同一个文件夹（train_images），并用一个txt文件（rec_gt_train.txt）记录图片路径和标签。
+```
+" 图像文件名                 图像标注信息 "
-**注意：** 默认请将图片路径和图片标签用 \t 分割，如用其他方式分割将造成训练报错
+train_data/rec/train/word_001.jpg   简单可依赖
+train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
+...
+```
+最终训练集应有如下文件结构：
+```
+|-train_data
+  |-rec
+    |- rec_gt_train.txt
+    |- train
+        |- word_001.png
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
 ```
-" 图像文件名                 图像标注信息 "
-train_data/train_0001.jpg   简单可依赖
+- 测试集
-train_data/train_0002.jpg   用科技让复杂的世界更简单
+同训练集类似，测试集也需要提供一个包含所有图片的文件夹（test）和一个rec_gt_test.txt，测试集的结构如下所示：
+```
+|-train_data
+  |-rec
+    |- rec_gt_test.txt
+    |- test
+        |- word_001.jpg
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
 ```
-PaddleOCR 提供了一份用于训练 icdar2015 数据集的标签文件，通过以下方式下载：
+<a name="数据下载"></a>
+1.2 数据下载
+若您本地没有数据集，可以在官网下载 [icdar2015](http://rrc.cvc.uab.es/?ch=4&com=downloads) 数据，用于快速验证。也可以参考[DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) ，下载 benchmark 所需的lmdb格式数据集。
+如果你使用的是icdar2015的公开数据集，PaddleOCR 提供了一份用于训练 icdar2015 数据集的标签文件，通过以下方式下载：
+如果希望复现SRN的论文指标，需要下载离线[增广数据](https://pan.baidu.com/s/1-HSZ-ZVdqBF2HaBZ5pRAKA),提取码: y3ry。增广数据是由MJSynth和SynthText做旋转和扰动得到的。数据下载完成后请解压到 {your_path}/PaddleOCR/train_data/data_lmdb_release/training/ 路径下。
 ```
 # 训练集标签
@@ -71,34 +104,8 @@ PaddleOCR 也提供了数据格式转换脚本，可以将官网 label 转换支
 python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
 ```
-最终训练集应有如下文件结构：
-```
-|-train_data
-    |-ic15_data
-        |- rec_gt_train.txt
-        |- train
-            |- word_001.png
-            |- word_002.jpg
-            |- word_003.jpg
-            | ...
-```
- 测试集
-同训练集类似，测试集也需要提供一个包含所有图片的文件夹（test）和一个rec_gt_test.txt，测试集的结构如下所示：
-```
-|-train_data
-    |-ic15_data
-        |- rec_gt_test.txt
-        |- test
-            |- word_001.jpg
-            |- word_002.jpg
-            |- word_003.jpg
-            | ...
-```
 <a name="字典"></a>
- 字典
+1.3 字典
 最后需要提供一个字典（{word_dict_name}.txt），使模型在训练时，可以将所有出现的字符映射为字典的索引。
@@ -115,6 +122,10 @@ n
 word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，“and” 将被映射成 [2 5 1]
+* 内置字典
+PaddleOCR内置了一部分字典，可以按需使用。
 `ppocr/utils/ppocr_keys_v1.txt` 是一个包含6623个字符的中文字典
 `ppocr/utils/ic15_dict.txt` 是一个包含36个字符的英文字典
@@ -130,7 +141,7 @@ word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，
 `ppocr/utils/dict/en_dict.txt` 是一个包含63个字符的英文字典
-您可以按需使用。
 目前的多语言模型仍处在demo阶段，会持续优化模型并补充语种，**非常欢迎您为我们提供其他语言的字典和字体**，
 如您愿意可将字典文件提交至 [dict](../../ppocr/utils/dict)，我们会在Repo中感谢您。
@@ -141,13 +152,13 @@ word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，
 并将 `character_type` 设置为 `ch`。
 <a name="支持空格"></a>
- 添加空格类别
+1.4 添加空格类别
 如果希望支持识别"空格"类别, 请将yml文件中的 `use_space_char` 字段设置为 `True`。
 <a name="启动训练"></a>
-### 启动训练
+### 2. 启动训练
 PaddleOCR提供了训练脚本、评估脚本和预测脚本，本节将以 CRNN 识别模型为例：
@@ -172,7 +183,7 @@ tar -xf rec_mv3_none_bilstm_ctc_v2.0_train.tar && rm -rf rec_mv3_none_bilstm_ctc
 python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_icdar15_train.yml
 ```
 <a name="数据增强"></a>
- 数据增强
+#### 2.1 数据增强
 PaddleOCR提供了多种数据增强方式，如果您希望在训练时加入扰动，请在配置文件中设置 `distort: true`。
@@ -183,7 +194,7 @@ PaddleOCR提供了多种数据增强方式，如果您希望在训练时加入
 *由于OpenCV的兼容性问题，扰动操作暂时只支持Linux*
 <a name="训练"></a>
- 训练
+#### 2.2 训练
 PaddleOCR支持训练和评估交替进行, 可以在 `configs/rec/rec_icdar15_train.yml` 中修改 `eval_batch_step` 设置评估频率，默认每500个iter评估一次。评估过程中默认将最佳acc模型，保存为 `output/rec_CRNN/best_accuracy` 。
@@ -272,7 +283,7 @@ Eval:
 **注意，预测/评估时的配置文件请务必与训练一致。**
 <a name="小语种"></a>
- 小语种
+#### 2.3 小语种
 PaddleOCR目前已支持26种（除中文外）语种识别，`configs/rec/multi_languages` 路径下提供了一个多语言的配置文件模版: [rec_multi_language_lite_train.yml](../../configs/rec/multi_language/rec_multi_language_lite_train.yml)。
@@ -415,7 +426,7 @@ Eval:
    ...
 ```
 <a name="评估"></a>
-### 评估
+### 3 评估
 评估数据集可以通过 `configs/rec/rec_icdar15_train.yml`  修改Eval中的 `label_file_path` 设置。
@@ -425,10 +436,10 @@ python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec
 ```
 <a name="预测"></a>
-### 预测
+### 4 预测
 <a name="训练引擎预测"></a>
-* 训练引擎的预测
+#### 4.1 训练引擎的预测
 使用 PaddleOCR 训练好的模型，可以通过以下脚本进行快速预测。

--- a/doc/doc_ch/whl.md
+++ b/doc/doc_ch/whl.md
 # paddleocr package使用说明
-## 快速上手
+## 1 快速上手
-### 安装whl包
+### 1.1 安装whl包
 pip安装
 ```bash
@@ -14,9 +14,12 @@ pip install "paddleocr>=2.0.1" # 推荐使用2.0.1+版本
 python3 setup.py bdist_wheel
 pip3 install dist/paddleocr-x.x.x-py3-none-any.whl # x.x.x是paddleocr的版本号
 ```
-### 1. 代码使用
-* 检测+分类+识别全流程
+## 2 使用
+### 2.1 代码使用
+paddleocr whl包会自动下载ppocr轻量级模型作为默认模型，可以根据第3节**自定义模型**进行自定义更换。
+* 检测+方向分类器+识别全流程
 ```python
 from paddleocr import PaddleOCR, draw_ocr
 # Paddleocr目前支持中英文、英文、法语、德语、韩语、日语，可以通过修改lang参数进行切换
@@ -33,7 +36,7 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
@@ -66,7 +69,7 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
@@ -84,7 +87,7 @@ im_show.save('result.jpg')
 </div>
-* 分类+识别
+* 方向分类器+识别
 ```python
 from paddleocr import PaddleOCR
 ocr = PaddleOCR(use_angle_cls=True) # need to run only once to download and load model into memory
@@ -111,7 +114,7 @@ for line in result:
 from PIL import Image
 image = Image.open(img_path).convert('RGB')
-im_show = draw_ocr(image, result, txts=None, scores=None, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, result, txts=None, scores=None, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
@@ -143,7 +146,7 @@ for line in result:
 ['韩国小馆', 0.9907421]
 ```
-* 单独执行分类
+* 单独执行方向分类器
 ```python
 from paddleocr import PaddleOCR
 ocr = PaddleOCR(use_angle_cls=True) # need to run only once to download and load model into memory
@@ -157,14 +160,14 @@ for line in result:
 ['0', 0.9999924]
 ```
-### 通过命令行使用
+### 2.2 通过命令行使用
 查看帮助信息
 ```bash
 paddleocr -h
 ```
-* 检测+分类+识别全流程
+* 检测+方向分类器+识别全流程
 ```bash
 paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --use_angle_cls true
 ```
@@ -188,7 +191,7 @@ paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg
 ......
 ```
-* 分类+识别
+* 方向分类器+识别
 ```bash
 paddleocr --image_dir PaddleOCR/doc/imgs_words/ch/word_1.jpg --use_angle_cls true --det false
 ```
@@ -220,7 +223,7 @@ paddleocr --image_dir PaddleOCR/doc/imgs_words/ch/word_1.jpg --det false
 ['韩国小馆', 0.9907421]
 ```
-* 单独执行分类
+* 单独执行方向分类器
 ```bash
 paddleocr --image_dir PaddleOCR/doc/imgs_words/ch/word_1.jpg --use_angle_cls true --det false --rec false
 ```
@@ -230,11 +233,11 @@ paddleocr --image_dir PaddleOCR/doc/imgs_words/ch/word_1.jpg --use_angle_cls tru
 ['0', 0.9999924]
 ```
-## 自定义模型
+## 3 自定义模型
 当内置模型无法满足需求时，需要使用到自己训练的模型。
 首先，参照[inference.md](./inference.md) 第一节转换将检测、分类和识别模型转换为inference模型，然后按照如下方式使用
-### 代码使用
+### 3.1 代码使用
 ```python
 from paddleocr import PaddleOCR, draw_ocr
 # 模型路径下必须含有model和params文件
@@ -250,22 +253,22 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
-### 通过命令行使用
+### 3.2 通过命令行使用
 ```bash
 paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --det_model_dir {your_det_model_dir} --rec_model_dir {your_rec_model_dir} --rec_char_dict_path {your_rec_char_dict_path} --cls_model_dir {your_cls_model_dir} --use_angle_cls true
 ```
-### 使用网络图片或者numpy数组作为输入
+## 4 使用网络图片或者numpy数组作为输入
-1. 网络图片
+### 4.1 网络图片
-代码使用
+- 代码使用
 ```python
 from paddleocr import PaddleOCR, draw_ocr
 # Paddleocr目前支持中英文、英文、法语、德语、韩语、日语，可以通过修改lang参数进行切换
@@ -282,16 +285,16 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
-命令行模式
+- 命令行模式
 ```bash
 paddleocr --image_dir http://n.sinaimg.cn/ent/transform/w630h933/20171222/o111-fypvuqf1838418.jpg --use_angle_cls=true
 ```
-2. numpy数组
+### 4.2 numpy数组
 仅通过代码使用时支持numpy数组作为输入
 ```python
 from paddleocr import PaddleOCR, draw_ocr
@@ -301,7 +304,7 @@ ocr = PaddleOCR(use_angle_cls=True, lang="ch") # need to run only once to downlo
 img_path = 'PaddleOCR/doc/imgs/11.jpg'
 img = cv2.imread(img_path)
 # img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY), 如果你自己训练的模型支持灰度图，可以将这句话的注释取消
-result = ocr.ocr(img_path, cls=True)
+result = ocr.ocr(img, cls=True)
 for line in result:
    print(line)
@@ -311,12 +314,12 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
-## 参数说明
+## 5 参数说明
 | 字段                    | 说明                                                                                                                                                                                                                 | 默认值                  |
 |-------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------|

--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -31,7 +31,9 @@ On Total-Text dataset, the text detection result is as follows:
 | --- | --- | --- | --- | --- | --- |
 |SAST|ResNet50_vd|89.63%|78.44%|83.66%|[Download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_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).
+**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)
 For the training guide and use of PaddleOCR text detection algorithms, please refer to the document [Text detection model training/evaluation/prediction](./detection_en.md)

--- a/doc/doc_en/angle_class_en.md
+++ b/doc/doc_en/angle_class_en.md
 ## TEXT ANGLE CLASSIFICATION
+### Method introduction
+The angle classification is used in the scene where the image is not 0 degrees. In this scene, it is necessary to perform a correction operation on the text line detected in the picture. In the PaddleOCR system,
+The text line image obtained after text detection is sent to the recognition model after affine transformation. At this time, only a 0 and 180 degree angle classification of the text is required, so the built-in PaddleOCR text angle classifier **only supports 0 and 180 degree classification**. If you want to support more angles, you can modify the algorithm yourself to support.
+Example of 0 and 180 degree data samples：
+![](../imgs_results/angle_class_example.jpg)
 ### DATA PREPARATION
 Please organize the dataset as follows:

--- a/doc/doc_en/inference_en.md
+++ b/doc/doc_en/inference_en.md
@@ -5,7 +5,8 @@ The inference model (the model saved by `paddle.jit.save`) is generally a solidi
 The model saved during the training process is the checkpoints model, which saves the parameters of the model and is mostly used to resume training.
-Compared with the checkpoints model, the inference model will additionally save the structural information of the model. It has superior performance in predicting in deployment and accelerating inferencing, is flexible and convenient, and is suitable for integration with actual systems. For more details, please refer to the document [Classification Framework](https://github.com/PaddlePaddle/PaddleClas/blob/master/docs/zh_CN/extension/paddle_inference.md).
+Compared with the checkpoints model, the inference model will additionally save the structural information of the model. Therefore, it is easier to deploy because the model structure and model parameters are already solidified in the inference model file, and is suitable for integration with actual systems.
+For more details, please refer to the document [Classification Framework](https://github.com/PaddlePaddle/PaddleClas/blob/release%2F2.0/docs/zh_CN/extension/paddle_mobile_inference.md).
 Next, we first introduce how to convert a trained model into an inference model, and then we will introduce text detection, text recognition, angle class, and the concatenation of them based on inference model.
@@ -147,7 +148,7 @@ The visual text detection results are saved to the ./inference_results folder by
 ![](../imgs_results/det_res_00018069.jpg)
 You can use the parameters `limit_type` and `det_limit_side_len` to limit the size of the input image,
-The optional parameters of `litmit_type` are [`max`, `min`], and
+The optional parameters of `limit_type` are [`max`, `min`], and
 `det_limit_size_len` is a positive integer, generally set to a multiple of 32, such as 960.
 The default setting of the parameters is `limit_type='max', det_limit_side_len=960`. Indicates that the longest side of the network input image cannot exceed 960,

--- a/doc/doc_en/installation_en.md
+++ b/doc/doc_en/installation_en.md
@@ -33,7 +33,7 @@ You can also visit [DockerHub](https://hub.docker.com/r/paddlepaddle/paddle/tags
 sudo docker container exec -it ppocr /bin/bash
 ```
-**2. Install PaddlePaddle Fluid v2.0**
+**2. Install PaddlePaddle 2.0**
 ```
 pip3 install --upgrade pip

--- a/doc/doc_en/recognition_en.md
+++ b/doc/doc_en/recognition_en.md
 ## TEXT RECOGNITION
- [DATA PREPARATION](#DATA_PREPARATION)
+- [1 DATA PREPARATION](#DATA_PREPARATION)
-    - [Dataset Download](#Dataset_download)
+    - [1.1 Costom Dataset](#Costom_Dataset)
-    - [Costom Dataset](#Costom_Dataset)  
+    - [1.2 Dataset Download](#Dataset_download)
-    - [Dictionary](#Dictionary)  
+    - [1.3 Dictionary](#Dictionary)  
-    - [Add Space Category](#Add_space_category)
+    - [1.4 Add Space Category](#Add_space_category)
- [TRAINING](#TRAINING)
+- [2 TRAINING](#TRAINING)
-    - [Data Augmentation](#Data_Augmentation)
+    - [2.1 Data Augmentation](#Data_Augmentation)
-    - [Training](#Training)
+    - [2.2 Training](#Training)
-    - [Multi-language](#Multi_language)
+    - [2.3 Multi-language](#Multi_language)
- [EVALUATION](#EVALUATION)
+- [3 EVALUATION](#EVALUATION)
- [PREDICTION](#PREDICTION)
+- [4 PREDICTION](#PREDICTION)
-    - [Training engine prediction](#Training_engine_prediction)
+    - [4.1 Training engine prediction](#Training_engine_prediction)
 <a name="DATA_PREPARATION"></a>
 ### DATA PREPARATION
-PaddleOCR supports two data formats: `LMDB` is used to train public data and evaluation algorithms; `general data` is used to train your own data:
+PaddleOCR supports two data formats:
+- `LMDB` is used to train data sets stored in lmdb format;
+- `general data` is used to train data sets stored in text files:
 Please organize the dataset as follows:
 The default storage path for training data is `PaddleOCR/train_data`, if you already have a dataset on your disk, just create a soft link to the dataset directory:
 ```
+# linux and mac os
 ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
 ```
-<a name="Dataset_download"></a>
-* Dataset download
-If you do not have a dataset locally, you can download it on the official website [icdar2015](http://rrc.cvc.uab.es/?ch=4&com=downloads). Also refer to [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here)，download the lmdb format dataset required for benchmark
-If you want to reproduce the paper indicators of SRN, you need to download offline [augmented data](https://pan.baidu.com/s/1-HSZ-ZVdqBF2HaBZ5pRAKA), extraction code: y3ry. The augmented data is obtained by rotation and perturbation of mjsynth and synthtext. Please unzip the data to {your_path}/PaddleOCR/train_data/data_lmdb_Release/training/path.
 <a name="Costom_Dataset"></a>
-* Use your own dataset:
+#### 1.1 Costom dataset
 If you want to use your own data for training, please refer to the following to organize your data.
 - Training set
-First put the training images in the same folder (train_images), and use a txt file (rec_gt_train.txt) to store the image path and label.
+It is recommended to put the training images in the same folder, and use a txt file (rec_gt_train.txt) to store the image path and label. The contents of the txt file are as follows:
 * Note: by default, the image path and image label are split with \t, if you use other methods to split, it will cause training error
 ```
 " Image file name           Image annotation "
-train_data/train_0001.jpg   简单可依赖
+train_data/rec/train/word_001.jpg   简单可依赖
-train_data/train_0002.jpg   用科技让复杂的世界更简单
+train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
-```
+...
-PaddleOCR provides label files for training the icdar2015 dataset, which can be downloaded in the following ways:
-```
-# Training set label
-wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt
-# Test Set Label
-wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt
 ```
 The final training set should have the following file structure:
 ```
 |-train_data
-    |-ic15_data
+  |-rec
-        |- rec_gt_train.txt
+    |- rec_gt_train.txt
-        |- train
+    |- train
-            |- word_001.png
+        |- word_001.png
-            |- word_002.jpg
+        |- word_002.jpg
-            |- word_003.jpg
+        |- word_003.jpg
-            | ...
+        | ...
 ```
 - Test set
@@ -82,6 +73,7 @@ Similar to the training set, the test set also needs to be provided a folder con
 ```
 |-train_data
+  |-rec
    |-ic15_data
        |- rec_gt_test.txt
        |- test
@@ -90,8 +82,25 @@ Similar to the training set, the test set also needs to be provided a folder con
            |- word_003.jpg
            | ...
 ```
+<a name="Dataset_download"></a>
+#### 1.2 Dataset download
+If you do not have a dataset locally, you can download it on the official website [icdar2015](http://rrc.cvc.uab.es/?ch=4&com=downloads). Also refer to [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) ，download the lmdb format dataset required for benchmark
+If you want to reproduce the paper indicators of SRN, you need to download offline [augmented data](https://pan.baidu.com/s/1-HSZ-ZVdqBF2HaBZ5pRAKA), extraction code: y3ry. The augmented data is obtained by rotation and perturbation of mjsynth and synthtext. Please unzip the data to {your_path}/PaddleOCR/train_data/data_lmdb_Release/training/path.
+PaddleOCR provides label files for training the icdar2015 dataset, which can be downloaded in the following ways:
+```
+# Training set label
+wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt
+# Test Set Label
+wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt
+```
 <a name="Dictionary"></a>
- Dictionary
+#### 1.3 Dictionary
 Finally, a dictionary ({word_dict_name}.txt) needs to be provided so that when the model is trained, all the characters that appear can be mapped to the dictionary index.
@@ -108,6 +117,8 @@ n
 In `word_dict.txt`, there is a single word in each line, which maps characters and numeric indexes together, e.g "and" will be mapped to [2 5 1]
+PaddleOCR has built-in dictionaries, which can be used on demand.
 `ppocr/utils/ppocr_keys_v1.txt` is a Chinese dictionary with 6623 characters.
 `ppocr/utils/ic15_dict.txt` is an English dictionary with 63 characters
@@ -123,8 +134,6 @@ In `word_dict.txt`, there is a single word in each line, which maps characters a
 `ppocr/utils/dict/en_dict.txt` is a English dictionary with 63 characters
-You can use it on demand.
 The current multi-language model is still in the demo stage and will continue to optimize the model and add languages. **You are very welcome to provide us with dictionaries and fonts in other languages**,
 If you like, you can submit the dictionary file to [dict](../../ppocr/utils/dict) and we will thank you in the Repo.
@@ -136,14 +145,14 @@ To customize the dict file, please modify the `character_dict_path` field in `co
 If you need to customize dic file, please add character_dict_path field in configs/rec/rec_icdar15_train.yml to point to your dictionary path. And set character_type to ch.
 <a name="Add_space_category"></a>
- Add space category
+#### 1.4 Add space category
 If you want to support the recognition of the `space` category, please set the `use_space_char` field in the yml file to `True`.
 **Note: use_space_char only takes effect when character_type=ch**
 <a name="TRAINING"></a>
-### TRAINING
+### 2 TRAINING
 PaddleOCR provides training scripts, evaluation scripts, and prediction scripts. In this section, the CRNN recognition model will be used as an example:
@@ -166,7 +175,7 @@ Start training:
 python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_icdar15_train.yml
 ```
 <a name="Data_Augmentation"></a>
- Data Augmentation
+#### 2.1 Data Augmentation
 PaddleOCR provides a variety of data augmentation methods. If you want to add disturbance during training, please set `distort: true` in the configuration file.
@@ -175,7 +184,7 @@ The default perturbation methods are: cvtColor, blur, jitter, Gasuss noise, rand
 Each disturbance method is selected with a 50% probability during the training process. For specific code implementation, please refer to: [img_tools.py](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/ppocr/data/rec/img_tools.py)
 <a name="Training"></a>
- Training
+#### 2.2 Training
 PaddleOCR supports alternating training and evaluation. You can modify `eval_batch_step` in `configs/rec/rec_icdar15_train.yml` to set the evaluation frequency. By default, it is evaluated every 500 iter and the best acc model is saved under `output/rec_CRNN/best_accuracy` during the evaluation process.
@@ -268,7 +277,7 @@ Eval:
 **Note that the configuration file for prediction/evaluation must be consistent with the training.**
 <a name="Multi_language"></a>
- Multi-language
+#### 2.3 Multi-language
 PaddleOCR currently supports 26 (except Chinese) language recognition. A multi-language configuration file template is
 provided under the path `configs/rec/multi_languages`: [rec_multi_language_lite_train.yml](../../configs/rec/multi_language/rec_multi_language_lite_train.yml)。
@@ -420,7 +429,7 @@ Eval:
 ```
 <a name="EVALUATION"></a>
-### EVALUATION
+### 3 EVALUATION
 The evaluation dataset can be set by modifying the `Eval.dataset.label_file_list` field in the `configs/rec/rec_icdar15_train.yml` file.
@@ -430,10 +439,10 @@ python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec
 ```
 <a name="PREDICTION"></a>
-### PREDICTION
+### 4 PREDICTION
 <a name="Training_engine_prediction"></a>
-* Training engine prediction
+#### 4.1 Training engine prediction
 Using the model trained by paddleocr, you can quickly get prediction through the following script.

--- a/doc/doc_en/whl_en.md
+++ b/doc/doc_en/whl_en.md
 # paddleocr package
-## Get started quickly
+## 1 Get started quickly
-### install package
+### 1.1 install package
 install by pypi
 ```bash
 pip install "paddleocr>=2.0.1" # Recommend to use version 2.0.1+
@@ -12,9 +12,11 @@ build own whl package and install
 python3 setup.py bdist_wheel
 pip3 install dist/paddleocr-x.x.x-py3-none-any.whl # x.x.x is the version of paddleocr
 ```
-### 1. Use by code
+## 2 Use
+### 2.1 Use by code
+The paddleocr whl package will automatically download the ppocr lightweight model as the default model, which can be customized and replaced according to the section 3 **Custom Model**.
-* detection classification and recognition
+* detection angle classification and recognition
 ```python
 from paddleocr import PaddleOCR,draw_ocr
 # Paddleocr supports Chinese, English, French, German, Korean and Japanese.
@@ -33,7 +35,7 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
@@ -67,7 +69,7 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
@@ -114,7 +116,7 @@ for line in result:
 from PIL import Image
 image = Image.open(img_path).convert('RGB')
-im_show = draw_ocr(image, result, txts=None, scores=None, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, result, txts=None, scores=None, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
@@ -163,7 +165,7 @@ Output will be a list, each item contains classification result and confidence
 ['0', 0.99999964]
 ```
-### Use by command line
+### 2.2 Use by command line
 show help information
 ```bash
@@ -239,11 +241,11 @@ Output will be a list, each item contains classification result and confidence
 ['0', 0.99999964]
 ```
-## Use custom model
+## 3 Use custom model
 When the built-in model cannot meet the needs, you need to use your own trained model.
 First, refer to the first section of [inference_en.md](./inference_en.md) to convert your det and rec model to inference model, and then use it as follows
-### 1. Use by code
+### 3.1 Use by code
 ```python
 from paddleocr import PaddleOCR,draw_ocr
@@ -260,22 +262,22 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
-### Use by command line
+### 3.2 Use by command line
 ```bash
 paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --det_model_dir {your_det_model_dir} --rec_model_dir {your_rec_model_dir} --rec_char_dict_path {your_rec_char_dict_path} --cls_model_dir {your_cls_model_dir} --use_angle_cls true
 ```
-### Use web images or numpy array as input
+## 4 Use web images or numpy array as input
-1. Web image
+### 4.1 Web image
-Use by code
+- Use by code
 ```python
 from paddleocr import PaddleOCR, draw_ocr
 ocr = PaddleOCR(use_angle_cls=True, lang="ch") # need to run only once to download and load model into memory
@@ -290,16 +292,16 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
-Use by command line
+- Use by command line
 ```bash
 paddleocr --image_dir http://n.sinaimg.cn/ent/transform/w630h933/20171222/o111-fypvuqf1838418.jpg --use_angle_cls=true
 ```
-2. Numpy array
+### 4.2 Numpy array
 Support numpy array as input only when used by code
 ```python
@@ -318,13 +320,13 @@ image = Image.open(img_path).convert('RGB')
 boxes = [line[0] for line in result]
 txts = [line[1][0] for line in result]
 scores = [line[1][1] for line in result]
-im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/simfang.ttf')
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
 im_show = Image.fromarray(im_show)
 im_show.save('result.jpg')
 ```
-## Parameter Description
+## 5 Parameter Description
 | Parameter                    | Description                                                                                                                                                                                                                 | Default value                  |
 |-------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------|

--- a/doc/imgs_results/angle_class_example.jpg
+++ b/doc/imgs_results/angle_class_example.jpg
--- a/doc/imgs_results/e2e_res_img293_pgnet.png
+++ b/doc/imgs_results/e2e_res_img293_pgnet.png