Merge branch 'dygraph' into update_whl

README.md

@@ -93,7 +93,7 @@ For a new language request, please refer to [Guideline for new language_requests
     - [Quick Inference Based on PIP](./doc/doc_en/whl_en.md)
     - [Python Inference](./doc/doc_en/inference_en.md)
     - [C++ Inference](./deploy/cpp_infer/readme_en.md)
-    - [Serving](./deploy/hubserving/readme_en.md)
+    - [Serving](./deploy/pdserving/README.md)
     - [Mobile](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/deploy/lite/readme_en.md)
     - [Benchmark](./doc/doc_en/benchmark_en.md)  
 - Data Annotation and Synthesis

README_ch.md

@@ -88,7 +88,7 @@ PaddleOCR同时支持动态图与静态图两种编程范式
     - [基于pip安装whl包快速推理](./doc/doc_ch/whl.md)
     - [基于Python脚本预测引擎推理](./doc/doc_ch/inference.md)
     - [基于C++预测引擎推理](./deploy/cpp_infer/readme.md)
-    - [服务化部署](./deploy/hubserving/readme.md)
+    - [服务化部署](./deploy/pdserving/README_CN.md)
     - [端侧部署](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/deploy/lite/readme.md)
     - [Benchmark](./doc/doc_ch/benchmark.md)
 - 数据集

deploy/hubserving/readme.md

@@ -2,7 +2,7 @@
 
 PaddleOCR提供2种服务部署方式：
 - 基于PaddleHub Serving的部署：代码路径为"`./deploy/hubserving`"，按照本教程使用；
-- (coming soon)基于PaddleServing的部署：代码路径为"`./deploy/pdserving`"，使用方法参考[文档](../../deploy/pdserving/readme.md)。
+- 基于PaddleServing的部署：代码路径为"`./deploy/pdserving`"，使用方法参考[文档](../../deploy/pdserving/README_CN.md)。
 
 # 基于PaddleHub Serving的服务部署
 

deploy/hubserving/readme_en.md

@@ -2,7 +2,7 @@ English | [简体中文](readme.md)
 
 PaddleOCR provides 2 service deployment methods:
 - Based on **PaddleHub Serving**: Code path is "`./deploy/hubserving`". Please follow this tutorial.
-- (coming soon)Based on **PaddleServing**: Code path is "`./deploy/pdserving`". Please refer to the [tutorial](../../deploy/pdserving/readme.md) for usage.
+- Based on **PaddleServing**: Code path is "`./deploy/pdserving`". Please refer to the [tutorial](../../deploy/pdserving/README.md) for usage.
 
 # Service deployment based on PaddleHub Serving  
 

deploy/pdserving/README.md

+# OCR Pipeline WebService
+
+(English|[简体中文](./README_CN.md))
+
+PaddleOCR provides two service deployment methods:
+- Based on **PaddleHub Serving**: Code path is "`./deploy/hubserving`". Please refer to the [tutorial](../../deploy/hubserving/readme_en.md)
+- Based on **PaddleServing**: Code path is "`./deploy/pdserving`". Please follow this tutorial.
+
+# Service deployment based on PaddleServing  
+
+This document will introduce how to use the [PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README.md) to deploy the PPOCR dynamic graph model as a pipeline online service.
+
+Some Key Features of Paddle Serving:
+- Integrate with Paddle training pipeline seamlessly, most paddle models can be deployed with one line command.
+- Industrial serving features supported, such as models management, online loading, online A/B testing etc.
+- Highly concurrent and efficient communication between clients and servers supported.
+
+The introduction and tutorial of Paddle Serving service deployment framework reference [document](https://github.com/PaddlePaddle/Serving/blob/develop/README.md).
+
+
+## Contents
+- [Environmental preparation](#environmental-preparation)
+- [Model conversion](#model-conversion)
+- [Paddle Serving pipeline deployment](#paddle-serving-pipeline-deployment)
+- [FAQ](#faq)
+
+<a name="environmental-preparation"></a>
+## Environmental preparation
+
+PaddleOCR operating environment and Paddle Serving operating environment are needed.
+
+1. Please prepare PaddleOCR operating environment reference [link](../../doc/doc_ch/installation.md).
+
+2. The steps of PaddleServing operating environment prepare are as follows:
+
+    Install serving which used to start the service
+    ```
+    pip3 install paddle-serving-server==0.5.0 # for CPU
+    pip3 install paddle-serving-server-gpu==0.5.0 # for GPU
+    # Other GPU environments need to confirm the environment and then choose to execute the following commands
+    pip3 install paddle-serving-server-gpu==0.5.0.post9 # GPU with CUDA9.0
+    pip3 install paddle-serving-server-gpu==0.5.0.post10 # GPU with CUDA10.0
+    pip3 install paddle-serving-server-gpu==0.5.0.post101 # GPU with CUDA10.1 + TensorRT6
+    pip3 install paddle-serving-server-gpu==0.5.0.post11 # GPU with CUDA10.1 + TensorRT7
+    ```
+
+3. Install the client to send requests to the service
+    ```
+    pip3 install paddle-serving-client==0.5.0 # for CPU
+
+    pip3 install paddle-serving-client-gpu==0.5.0 # for GPU
+    ```
+
+4. Install serving-app
+    ```
+    pip3 install paddle-serving-app==0.3.0
+    # fix local_predict to support load dynamic model
+    # find the install directoory of paddle_serving_app
+    vim /usr/local/lib/python3.7/site-packages/paddle_serving_app/local_predict.py
+    # replace line 85 of local_predict.py config = AnalysisConfig(model_path) with:
+    if os.path.exists(os.path.join(model_path, "__params__")):
+        config = AnalysisConfig(os.path.join(model_path, "__model__"), os.path.join(model_path, "__params__"))
+    else:
+        config = AnalysisConfig(model_path)
+    ```
+
+   **note:** If you want to install the latest version of PaddleServing, refer to [link](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md).
+
+
+<a name="model-conversion"></a>
+## Model conversion
+When using PaddleServing for service deployment, you need to convert the saved inference model into a serving model that is easy to deploy.
+
+Firstly, download the [inference model](https://github.com/PaddlePaddle/PaddleOCR#pp-ocr-20-series-model-listupdate-on-dec-15) of PPOCR
+```
+# Download and unzip the OCR text detection model
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_det_infer.tar
+# Download and unzip the OCR text recognition model
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar
+
+```
+Then, you can use installed paddle_serving_client tool to convert inference model to server model.
+```
+#  Detection model conversion
+python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_det_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_det_server_2.0_serving/ \
+                                         --serving_client ./ppocr_det_server_2.0_client/
+
+#  Recognition model conversion
+python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_rec_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_rec_server_2.0_serving/  \
+                                         --serving_client ./ppocr_rec_server_2.0_client/
+
+```
+
+After the detection model is converted, there will be additional folders of `ppocr_det_server_2.0_serving` and `ppocr_det_server_2.0_client` in the current folder, with the following format:
+```
+|- ppocr_det_server_2.0_serving/
+   |- __model__
+   |- __params__
+   |- serving_server_conf.prototxt
+   |- serving_server_conf.stream.prototxt
+
+|- ppocr_det_server_2.0_client
+   |- serving_client_conf.prototxt
+   |- serving_client_conf.stream.prototxt
+
+```
+The recognition model is the same.
+
+<a name="paddle-serving-pipeline-deployment"></a>
+## Paddle Serving pipeline deployment
+
+1. Download the PaddleOCR code, if you have already downloaded it, you can skip this step.
+    ```
+    git clone https://github.com/PaddlePaddle/PaddleOCR
+
+    # Enter the working directory  
+    cd PaddleOCR/deploy/pdserver/
+    ```
+
+    The pdserver directory contains the code to start the pipeline service and send prediction requests, including:
+    ```
+    __init__.py
+    config.yml # Start the service configuration file
+    ocr_reader.py # OCR model pre-processing and post-processing code implementation
+    pipeline_http_client.py # Script to send pipeline prediction request
+    web_service.py # Start the script of the pipeline server
+    ```
+
+2. Run the following command to start the service.
+    ```
+    # Start the service and save the running log in log.txt
+    python3 web_service.py &>log.txt &
+    ```
+    After the service is successfully started, a log similar to the following will be printed in log.txt
+    ![](./imgs/start_server.png)
+
+3. Send service request
+    ```
+    python3 pipeline_http_client.py
+    ```
+    After successfully running, the predicted result of the model will be printed in the cmd window. An example of the result is:
+    ![](./imgs/results.png)  
+
+<a name="faq"></a>
+## FAQ
+**Q1**: No result return after sending the request.
+
+**A1**: Do not set the proxy when starting the service and sending the request. You can close the proxy before starting the service and before sending the request. The command to close the proxy is:
+```
+unset https_proxy
+unset http_proxy
+```  

deploy/pdserving/README_CN.md

+# PPOCR 服务化部署
+
+([English](./README.md)|简体中文)
+
+PaddleOCR提供2种服务部署方式：
+- 基于PaddleHub Serving的部署：代码路径为"`./deploy/hubserving`"，使用方法参考[文档](../../deploy/hubserving/readme.md)；
+- 基于PaddleServing的部署：代码路径为"`./deploy/pdserving`"，按照本教程使用。
+
+# 基于PaddleServing的服务部署
+
+本文档将介绍如何使用[PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)工具部署PPOCR
+动态图模型的pipeline在线服务。
+
+相比较于hubserving部署，PaddleServing具备以下优点：
+- 支持客户端和服务端之间高并发和高效通信
+- 支持 工业级的服务能力 例如模型管理，在线加载，在线A/B测试等
+- 支持 多种编程语言 开发客户端，例如C++, Python和Java
+
+更多有关PaddleServing服务化部署框架介绍和使用教程参考[文档](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)。
+
+## 目录
+- [环境准备](#环境准备)
+- [模型转换](#模型转换)
+- [Paddle Serving pipeline部署](#部署)
+- [FAQ](#FAQ)
+
+<a name="环境准备"></a>
+## 环境准备
+
+需要准备PaddleOCR的运行环境和Paddle Serving的运行环境。
+
+- 准备PaddleOCR的运行环境参考[链接](../../doc/doc_ch/installation.md)
+
+- 准备PaddleServing的运行环境，步骤如下
+
+1. 安装serving，用于启动服务
+    ```
+    pip3 install paddle-serving-server==0.5.0 # for CPU
+    pip3 install paddle-serving-server-gpu==0.5.0 # for GPU
+    # 其他GPU环境需要确认环境再选择执行如下命令
+    pip3 install paddle-serving-server-gpu==0.5.0.post9 # GPU with CUDA9.0
+    pip3 install paddle-serving-server-gpu==0.5.0.post10 # GPU with CUDA10.0
+    pip3 install paddle-serving-server-gpu==0.5.0.post101 # GPU with CUDA10.1 + TensorRT6
+    pip3 install paddle-serving-server-gpu==0.5.0.post11 # GPU with CUDA10.1 + TensorRT7
+    ```
+
+2. 安装client，用于向服务发送请求
+    ```
+    pip3 install paddle-serving-client==0.5.0  # for CPU
+
+    pip3 install paddle-serving-client-gpu==0.5.0   # for GPU
+    ```
+
+3. 安装serving-app
+    ```
+    pip3 install paddle-serving-app==0.3.0
+    ```
+    **note:**  安装0.3.0版本的serving-app后，为了能加载动态图模型，需要修改serving_app的源码，具体为：
+    ```
+    # 找到paddle_serving_app的安装目录，找到并编辑local_predict.py文件
+    vim /usr/local/lib/python3.7/site-packages/paddle_serving_app/local_predict.py
+    # 将local_predict.py 的第85行 config = AnalysisConfig(model_path)  替换为：
+    if os.path.exists(os.path.join(model_path, "__params__")):
+        config = AnalysisConfig(os.path.join(model_path, "__model__"), os.path.join(model_path, "__params__"))
+    else:
+        config = AnalysisConfig(model_path)
+    ```
+
+    **Note:** 如果要安装最新版本的PaddleServing参考[链接](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md)。
+
+<a name="模型转换"></a>
+## 模型转换
+
+使用PaddleServing做服务化部署时，需要将保存的inference模型转换为serving易于部署的模型。
+
+首先，下载PPOCR的[inference模型](https://github.com/PaddlePaddle/PaddleOCR#pp-ocr-20-series-model-listupdate-on-dec-15)
+```
+# 下载并解压 OCR 文本检测模型
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_det_infer.tar
+# 下载并解压 OCR 文本识别模型
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar
+```
+
+接下来，用安装的paddle_serving_client把下载的inference模型转换成易于server部署的模型格式。
+
+```
+# 转换检测模型
+python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_det_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_det_server_2.0_serving/ \
+                                         --serving_client ./ppocr_det_server_2.0_client/
+
+# 转换识别模型
+python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_rec_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_rec_server_2.0_serving/  \
+                                         --serving_client ./ppocr_rec_server_2.0_client/
+```
+
+检测模型转换完成后，会在当前文件夹多出`ppocr_det_server_2.0_serving` 和`ppocr_det_server_2.0_client`的文件夹，具备如下格式：
+```
+|- ppocr_det_server_2.0_serving/
+  |- __model__  
+  |- __params__
+  |- serving_server_conf.prototxt  
+  |- serving_server_conf.stream.prototxt
+
+|- ppocr_det_server_2.0_client
+  |- serving_client_conf.prototxt  
+  |- serving_client_conf.stream.prototxt
+
+```
+识别模型同理。
+
+<a name="部署"></a>
+## Paddle Serving pipeline部署
+
+1. 下载PaddleOCR代码，若已下载可跳过此步骤
+    ```
+    git clone https://github.com/PaddlePaddle/PaddleOCR
+
+    # 进入到工作目录
+    cd PaddleOCR/deploy/pdserver/
+    ```
+    pdserver目录包含启动pipeline服务和发送预测请求的代码，包括：
+    ```
+    __init__.py
+    config.yml            # 启动服务的配置文件
+    ocr_reader.py         # OCR模型预处理和后处理的代码实现
+    pipeline_http_client.py   # 发送pipeline预测请求的脚本
+    web_service.py        # 启动pipeline服务端的脚本
+    ```
+
+2. 启动服务可运行如下命令：
+    ```
+    # 启动服务，运行日志保存在log.txt
+    python3 web_service.py &>log.txt &
+    ```
+    成功启动服务后，log.txt中会打印类似如下日志
+    ![](./imgs/start_server.png)
+
+3. 发送服务请求：
+    ```
+    python3 pipeline_http_client.py
+    ```
+    成功运行后，模型预测的结果会打印在cmd窗口中，结果示例为：
+    ![](./imgs/results.png)
+
+
+<a name="FAQ"></a>
+## FAQ
+**Q1**： 发送请求后没有结果返回或者提示输出解码报错
+
+**A1**： 启动服务和发送请求时不要设置代理，可以在启动服务前和发送请求前关闭代理，关闭代理的命令是：
+```
+unset https_proxy
+unset http_proxy
+```

deploy/pdserving/__init__.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.

deploy/pdserving/config.yml

+#rpc端口, rpc_port和http_port不允许同时为空。当rpc_port为空且http_port不为空时，会自动将rpc_port设置为http_port+1
+rpc_port: 18090
+
+#http端口, rpc_port和http_port不允许同时为空。当rpc_port可用且http_port为空时，不自动生成http_port
+http_port: 9999
+
+#worker_num, 最大并发数。当build_dag_each_worker=True时, 框架会创建worker_num个进程，每个进程内构建grpcSever和DAG
+##当build_dag_each_worker=False时，框架会设置主线程grpc线程池的max_workers=worker_num
+worker_num: 20
+
+#build_dag_each_worker, False，框架在进程内创建一条DAG；True，框架会每个进程内创建多个独立的DAG
+build_dag_each_worker: false
+
+dag:
+    #op资源类型, True, 为线程模型；False，为进程模型
+    is_thread_op: False
+
+    #重试次数
+    retry: 1
+
+    #使用性能分析, True，生成Timeline性能数据，对性能有一定影响；False为不使用
+    use_profile: False
+    
+    tracer:
+        interval_s: 10
+op:
+    det:
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 4
+
+        #当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
+        local_service_conf:
+            #client类型，包括brpc, grpc和local_predictor.local_predictor不启动Serving服务，进程内预测
+            client_type: local_predictor
+
+            #det模型路径
+            model_config: /paddle/serving/models/det_serving_server/ #ocr_det_model
+
+            #Fetch结果列表，以client_config中fetch_var的alias_name为准
+            fetch_list: ["save_infer_model/scale_0.tmp_1"]
+
+            #计算硬件ID，当devices为""或不写时为CPU预测；当devices为"0", "0,1,2"时为GPU预测，表示使用的GPU卡
+            devices: "2"
+            
+            ir_optim: True
+    rec:
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 1
+
+        #超时时间, 单位ms
+        timeout: -1
+ 
+        #Serving交互重试次数，默认不重试
+        retry: 1
+
+        #当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
+        local_service_conf:
+
+            #client类型，包括brpc, grpc和local_predictor。local_predictor不启动Serving服务，进程内预测
+            client_type: local_predictor
+
+            #rec模型路径
+            model_config: /paddle/serving/models/rec_serving_server/ #ocr_rec_model
+
+            #Fetch结果列表，以client_config中fetch_var的alias_name为准
+            fetch_list: ["save_infer_model/scale_0.tmp_1"] #["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"] 
+
+            #计算硬件ID，当devices为""或不写时为CPU预测；当devices为"0", "0,1,2"时为GPU预测，表示使用的GPU卡
+            devices: "2"
+            
+            ir_optim: True

deploy/pdserving/ocr_reader.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.
+
+import cv2
+import copy
+import numpy as np
+import math
+import re
+import sys
+import argparse
+import string
+from copy import deepcopy
+import paddle
+
+
+class DetResizeForTest(object):
+    def __init__(self, **kwargs):
+        super(DetResizeForTest, self).__init__()
+        self.resize_type = 0
+        if 'image_shape' in kwargs:
+            self.image_shape = kwargs['image_shape']
+            self.resize_type = 1
+        elif 'limit_side_len' in kwargs:
+            self.limit_side_len = kwargs['limit_side_len']
+            self.limit_type = kwargs.get('limit_type', 'min')
+        elif 'resize_long' in kwargs:
+            self.resize_type = 2
+            self.resize_long = kwargs.get('resize_long', 960)
+        else:
+            self.limit_side_len = 736
+            self.limit_type = 'min'
+
+    def __call__(self, data):
+        img = deepcopy(data)
+        src_h, src_w, _ = img.shape
+
+        if self.resize_type == 0:
+            img, [ratio_h, ratio_w] = self.resize_image_type0(img)
+        elif self.resize_type == 2:
+            img, [ratio_h, ratio_w] = self.resize_image_type2(img)
+        else:
+            img, [ratio_h, ratio_w] = self.resize_image_type1(img)
+
+        return img
+
+    def resize_image_type1(self, img):
+        resize_h, resize_w = self.image_shape
+        ori_h, ori_w = img.shape[:2]  # (h, w, c)
+        ratio_h = float(resize_h) / ori_h
+        ratio_w = float(resize_w) / ori_w
+        img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        return img, [ratio_h, ratio_w]
+
+    def resize_image_type0(self, img):
+        """
+        resize image to a size multiple of 32 which is required by the network
+        args:
+            img(array): array with shape [h, w, c]
+        return(tuple):
+            img, (ratio_h, ratio_w)
+        """
+        limit_side_len = self.limit_side_len
+        h, w, _ = img.shape
+
+        # limit the max side
+        if self.limit_type == 'max':
+            if max(h, w) > limit_side_len:
+                if h > w:
+                    ratio = float(limit_side_len) / h
+                else:
+                    ratio = float(limit_side_len) / w
+            else:
+                ratio = 1.
+        else:
+            if min(h, w) < limit_side_len:
+                if h < w:
+                    ratio = float(limit_side_len) / h
+                else:
+                    ratio = float(limit_side_len) / w
+            else:
+                ratio = 1.
+        resize_h = int(h * ratio)
+        resize_w = int(w * ratio)
+
+        resize_h = int(round(resize_h / 32) * 32)
+        resize_w = int(round(resize_w / 32) * 32)
+
+        try:
+            if int(resize_w) <= 0 or int(resize_h) <= 0:
+                return None, (None, None)
+            img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        except:
+            print(img.shape, resize_w, resize_h)
+            sys.exit(0)
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+        # return img, np.array([h, w])
+        return img, [ratio_h, ratio_w]
+
+    def resize_image_type2(self, img):
+        h, w, _ = img.shape
+
+        resize_w = w
+        resize_h = h
+
+        # Fix the longer side
+        if resize_h > resize_w:
+            ratio = float(self.resize_long) / resize_h
+        else:
+            ratio = float(self.resize_long) / resize_w
+
+        resize_h = int(resize_h * ratio)
+        resize_w = int(resize_w * ratio)
+
+        max_stride = 128
+        resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+        resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+        img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+
+        return img, [ratio_h, ratio_w]
+
+
+class BaseRecLabelDecode(object):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, config):
+        support_character_type = [
+            'ch', 'en', 'EN_symbol', 'french', 'german', 'japan', 'korean',
+            'it', 'xi', 'pu', 'ru', 'ar', 'ta', 'ug', 'fa', 'ur', 'rs', 'oc',
+            'rsc', 'bg', 'uk', 'be', 'te', 'ka', 'chinese_cht', 'hi', 'mr',
+            'ne', 'EN'
+        ]
+        character_type = config['character_type']
+        character_dict_path = config['character_dict_path']
+        use_space_char = True
+        assert character_type in support_character_type, "Only {} are supported now but get {}".format(
+            support_character_type, character_type)
+
+        self.beg_str = "sos"
+        self.end_str = "eos"
+
+        if character_type == "en":
+            self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+            dict_character = list(self.character_str)
+        elif character_type == "EN_symbol":
+            # same with ASTER setting (use 94 char).
+            self.character_str = string.printable[:-6]
+            dict_character = list(self.character_str)
+        elif character_type in support_character_type:
+            self.character_str = ""
+            assert character_dict_path is not None, "character_dict_path should not be None when character_type is {}".format(
+                character_type)
+            with open(character_dict_path, "rb") as fin:
+                lines = fin.readlines()
+                for line in lines:
+                    line = line.decode('utf-8').strip("\n").strip("\r\n")
+                    self.character_str += line
+            if use_space_char:
+                self.character_str += " "
+            dict_character = list(self.character_str)
+
+        else:
+            raise NotImplementedError
+        self.character_type = character_type
+        dict_character = self.add_special_char(dict_character)
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.character = dict_character
+
+    def add_special_char(self, dict_character):
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """ convert text-index into text-label. """
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if is_remove_duplicate:
+                    # only for predict
+                    if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
+                            batch_idx][idx]:
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][
+                    idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = ''.join(char_list)
+            result_list.append((text, np.mean(conf_list)))
+        return result_list
+
+    def get_ignored_tokens(self):
+        return [0]  # for ctc blank
+
+
+class CTCLabelDecode(BaseRecLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(
+            self,
+            config,
+            #character_dict_path=None,
+            #character_type='ch',
+            #use_space_char=False,
+            **kwargs):
+        super(CTCLabelDecode, self).__init__(config)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['blank'] + dict_character
+        return dict_character
+
+
+class CharacterOps(object):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, config):
+        self.character_type = config['character_type']
+        self.loss_type = config['loss_type']
+        if self.character_type == "en":
+            self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+            dict_character = list(self.character_str)
+        elif self.character_type == "ch":
+            character_dict_path = config['character_dict_path']
+            self.character_str = ""
+            with open(character_dict_path, "rb") as fin:
+                lines = fin.readlines()
+                for line in lines:
+                    line = line.decode('utf-8').strip("\n").strip("\r\n")
+                    self.character_str += line
+            dict_character = list(self.character_str)
+        elif self.character_type == "en_sensitive":
+            # same with ASTER setting (use 94 char).
+            self.character_str = string.printable[:-6]
+            dict_character = list(self.character_str)
+        else:
+            self.character_str = None
+        assert self.character_str is not None, \
+            "Nonsupport type of the character: {}".format(self.character_str)
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        if self.loss_type == "attention":
+            dict_character = [self.beg_str, self.end_str] + dict_character
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.character = dict_character
+
+    def encode(self, text):
+        """convert text-label into text-index.
+        input:
+            text: text labels of each image. [batch_size]
+
+        output:
+            text: concatenated text index for CTCLoss.
+                    [sum(text_lengths)] = [text_index_0 + text_index_1 + ... + text_index_(n - 1)]
+            length: length of each text. [batch_size]
+        """
+        if self.character_type == "en":
+            text = text.lower()
+
+        text_list = []
+        for char in text:
+            if char not in self.dict:
+                continue
+            text_list.append(self.dict[char])
+        text = np.array(text_list)
+        return text
+
+    def decode(self, text_index, is_remove_duplicate=False):
+        """ convert text-index into text-label. """
+        char_list = []
+        char_num = self.get_char_num()
+
+        if self.loss_type == "attention":
+            beg_idx = self.get_beg_end_flag_idx("beg")
+            end_idx = self.get_beg_end_flag_idx("end")
+            ignored_tokens = [beg_idx, end_idx]
+        else:
+            ignored_tokens = [char_num]
+
+        for idx in range(len(text_index)):
+            if text_index[idx] in ignored_tokens:
+                continue
+            if is_remove_duplicate:
+                if idx > 0 and text_index[idx - 1] == text_index[idx]:
+                    continue
+            char_list.append(self.character[text_index[idx]])
+        text = ''.join(char_list)
+        return text
+
+    def get_char_num(self):
+        return len(self.character)
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if self.loss_type == "attention":
+            if beg_or_end == "beg":
+                idx = np.array(self.dict[self.beg_str])
+            elif beg_or_end == "end":
+                idx = np.array(self.dict[self.end_str])
+            else:
+                assert False, "Unsupport type %s in get_beg_end_flag_idx"\
+                    % beg_or_end
+            return idx
+        else:
+            err = "error in get_beg_end_flag_idx when using the loss %s"\
+                % (self.loss_type)
+            assert False, err
+
+
+class OCRReader(object):
+    def __init__(self,
+                 algorithm="CRNN",
+                 image_shape=[3, 32, 320],
+                 char_type="ch",
+                 batch_num=1,
+                 char_dict_path="./ppocr_keys_v1.txt"):
+        self.rec_image_shape = image_shape
+        self.character_type = char_type
+        self.rec_batch_num = batch_num
+        char_ops_params = {}
+        char_ops_params["character_type"] = char_type
+        char_ops_params["character_dict_path"] = char_dict_path
+        char_ops_params['loss_type'] = 'ctc'
+        self.char_ops = CharacterOps(char_ops_params)
+        self.label_ops = CTCLabelDecode(char_ops_params)
+
+    def resize_norm_img(self, img, max_wh_ratio):
+        imgC, imgH, imgW = self.rec_image_shape
+        if self.character_type == "ch":
+            imgW = int(32 * max_wh_ratio)
+        h = img.shape[0]
+        w = img.shape[1]
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+        resized_image = resized_image.astype('float32')
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+
+        padding_im[:, :, 0:resized_w] = resized_image
+        return padding_im
+
+    def preprocess(self, img_list):
+        img_num = len(img_list)
+        norm_img_batch = []
+        max_wh_ratio = 0
+        for ino in range(img_num):
+            h, w = img_list[ino].shape[0:2]
+            wh_ratio = w * 1.0 / h
+            max_wh_ratio = max(max_wh_ratio, wh_ratio)
+
+        for ino in range(img_num):
+            norm_img = self.resize_norm_img(img_list[ino], max_wh_ratio)
+            norm_img = norm_img[np.newaxis, :]
+            norm_img_batch.append(norm_img)
+        norm_img_batch = np.concatenate(norm_img_batch)
+        norm_img_batch = norm_img_batch.copy()
+
+        return norm_img_batch[0]
+
+    def postprocess_old(self, outputs, with_score=False):
+        rec_res = []
+        rec_idx_lod = outputs["ctc_greedy_decoder_0.tmp_0.lod"]
+        rec_idx_batch = outputs["ctc_greedy_decoder_0.tmp_0"]
+        if with_score:
+            predict_lod = outputs["softmax_0.tmp_0.lod"]
+        for rno in range(len(rec_idx_lod) - 1):
+            beg = rec_idx_lod[rno]
+            end = rec_idx_lod[rno + 1]
+            if isinstance(rec_idx_batch, list):
+                rec_idx_tmp = [x[0] for x in rec_idx_batch[beg:end]]
+            else:  #nd array
+                rec_idx_tmp = rec_idx_batch[beg:end, 0]
+            preds_text = self.char_ops.decode(rec_idx_tmp)
+            if with_score:
+                beg = predict_lod[rno]
+                end = predict_lod[rno + 1]
+                if isinstance(outputs["softmax_0.tmp_0"], list):
+                    outputs["softmax_0.tmp_0"] = np.array(outputs[
+                        "softmax_0.tmp_0"]).astype(np.float32)
+                probs = outputs["softmax_0.tmp_0"][beg:end, :]
+                ind = np.argmax(probs, axis=1)
+                blank = probs.shape[1]
+                valid_ind = np.where(ind != (blank - 1))[0]
+                score = np.mean(probs[valid_ind, ind[valid_ind]])
+                rec_res.append([preds_text, score])
+            else:
+                rec_res.append([preds_text])
+        return rec_res
+
+    def postprocess(self, outputs, with_score=False):
+        preds = outputs["save_infer_model/scale_0.tmp_1"]
+        try:
+            preds = preds.numpy()
+        except:
+            pass
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.label_ops.decode(
+            preds_idx, preds_prob, is_remove_duplicate=True)
+        return text

deploy/pdserving/pipeline_http_client.py

+# Copyright (c) 2020 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.
+
+import numpy as np
+import requests
+import json
+import base64
+import os
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+
+url = "http://127.0.0.1:9999/ocr/prediction"
+test_img_dir = "../doc/imgs/"
+for idx, img_file in enumerate(os.listdir(test_img_dir)):
+    with open(os.path.join(test_img_dir, img_file), 'rb') as file:
+        image_data1 = file.read()
+
+    image = cv2_to_base64(image_data1)
+
+    for i in range(1):
+        data = {"key": ["image"], "value": [image]}
+        r = requests.post(url=url, data=json.dumps(data))
+        print(r.json())
+
+test_img_dir = "../doc/imgs/"
+print("==> total number of test imgs: ", len(os.listdir(test_img_dir)))

deploy/pdserving/pipeline_rpc_client.py

+# Copyright (c) 2020 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.pipeline import PipelineClient
+except ImportError:
+    from paddle_serving_server.pipeline import PipelineClient
+import numpy as np
+import requests
+import json
+import cv2
+import base64
+import os
+
+client = PipelineClient()
+client.connect(['127.0.0.1:18090'])
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+
+test_img_dir = "imgs/"
+for img_file in os.listdir(test_img_dir):
+    with open(os.path.join(test_img_dir, img_file), 'rb') as file:
+        image_data = file.read()
+    image = cv2_to_base64(image_data)
+
+for i in range(1):
+    ret = client.predict(feed_dict={"image": image}, fetch=["res"])
+    print(ret)
+    #print(ret)

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()

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)。
 

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)
 

paddleocr.py

@@ -236,7 +236,9 @@ class PaddleOCR(predict_system.TextSystem):
         assert lang in model_urls[
             'rec'], 'param lang must in {}, but got {}'.format(
                 model_urls['rec'].keys(), lang)
+        use_inner_dict = False
         if postprocess_params.rec_char_dict_path is None:
+            use_inner_dict = True
             postprocess_params.rec_char_dict_path = model_urls['rec'][lang][
                 'dict_path']
 
@@ -263,7 +265,7 @@ class PaddleOCR(predict_system.TextSystem):
         if postprocess_params.rec_algorithm not in SUPPORT_REC_MODEL:
             logger.error('rec_algorithm must in {}'.format(SUPPORT_REC_MODEL))
             sys.exit(0)
-
+        if use_inner_dict:
             postprocess_params.rec_char_dict_path = str(
                 Path(__file__).parent / postprocess_params.rec_char_dict_path)
 
@@ -282,8 +284,13 @@ class PaddleOCR(predict_system.TextSystem):
         if isinstance(img, list) and det == True:
             logger.error('When input a list of images, det must be false')
             exit(0)
+        if cls == False:
+            self.use_angle_cls = False
+        elif cls == True and self.use_angle_cls == False:
+            logger.warning(
+                'Since the angle classifier is not initialized, the angle classifier will not be uesd during the forward process'
+            )
 
-        self.use_angle_cls = cls
         if isinstance(img, str):
             # download net image
             if img.startswith('http'):

ppocr/data/imaug/randaugment.py

@@ -117,13 +117,16 @@ class RawRandAugment(object):
 class RandAugment(RawRandAugment):
     """ RandAugment wrapper to auto fit different img types """
 
-    def __init__(self, *args, **kwargs):
+    def __init__(self, prob=0.5, *args, **kwargs):
+        self.prob = prob
         if six.PY2:
             super(RandAugment, self).__init__(*args, **kwargs)
         else:
             super().__init__(*args, **kwargs)
 
     def __call__(self, data):
+        if np.random.rand() > self.prob:
+            return data
         img = data['image']
         if not isinstance(img, Image.Image):
             img = np.ascontiguousarray(img)