Merge pull request #1054 from MissPenguin/develop

update pdserving

deploy/pdserving/clas_local_server.py

@@ -107,16 +107,18 @@ class OCRService(WebService):
             if ".lod" in x:
                 self.tmp_args[x] = fetch_map[x]
         _, rec_res = self.text_classifier.postprocess(outputs, self.tmp_args)
-        res = {
+        res = []
-            "pred_text": [x[0] for x in rec_res],
+        for i in range(len(rec_res)):
-            "score": [str(x[1]) for x in rec_res]
+            res.append({
-        }
+                "direction": rec_res[i][0],
+                "confidence": float(rec_res[i][1])
+            })
         return res
 
 
 if __name__ == "__main__":
     ocr_service = OCRService(name="ocr")
-    ocr_service.load_model_config(global_args.cls_model_dir)
+    ocr_service.load_model_config(global_args.cls_server_dir)
     ocr_service.init_rec()
     if global_args.use_gpu:
         ocr_service.prepare_server(

deploy/pdserving/clas_rpc_server.py

@@ -113,16 +113,18 @@ class OCRService(WebService):
             if ".lod" in x:
                 self.tmp_args[x] = fetch_map[x]
         _, rec_res = self.text_classifier.postprocess(outputs, self.tmp_args)
-        res = {
+        res = []
-            "direction": [x[0] for x in rec_res],
+        for i in range(len(rec_res)):
-            "score": [str(x[1]) for x in rec_res]
+            res.append({
-        }
+                "direction": rec_res[i][0],
+                "confidence": float(rec_res[i][1])
+            })
         return res
 
 
 if __name__ == "__main__":
     ocr_service = OCRService(name="ocr")
-    ocr_service.load_model_config(global_args.cls_model_dir)
+    ocr_service.load_model_config(global_args.cls_server_dir)
     ocr_service.init_rec()
     if global_args.use_gpu:
         ocr_service.prepare_server(

deploy/pdserving/det_local_server.py

@@ -90,13 +90,15 @@ class DetService(WebService):
 
     def postprocess(self, feed={}, fetch=[], fetch_map=None):
         outputs = [fetch_map[x] for x in fetch]
-        res = self.text_detector.postprocess(outputs, self.tmp_args)
+        det_res = self.text_detector.postprocess(outputs, self.tmp_args)
-        return {"boxes": res.tolist()}
+        res = []
-
+        for i in range(len(det_res)):
+            res.append({"text_region": det_res[i].tolist()})
+        return res
 
 if __name__ == "__main__":
     ocr_service = DetService(name="ocr")
-    ocr_service.load_model_config(global_args.det_model_dir)
+    ocr_service.load_model_config(global_args.det_server_dir)
     ocr_service.init_det()
     if global_args.use_gpu:
         ocr_service.prepare_server(

deploy/pdserving/det_rpc_server.py

@@ -89,13 +89,15 @@ class DetService(WebService):
 
     def postprocess(self, feed={}, fetch=[], fetch_map=None):
         outputs = [fetch_map[x] for x in fetch]
-        res = self.text_detector.postprocess(outputs, self.tmp_args)
+        det_res = self.text_detector.postprocess(outputs, self.tmp_args)
-        return {"boxes": res.tolist()}
+        res = []
-
+        for i in range(len(det_res)):
+            res.append({"text_region": det_res[i].tolist()})
+        return res
 
 if __name__ == "__main__":
     ocr_service = DetService(name="ocr")
-    ocr_service.load_model_config(global_args.det_model_dir)
+    ocr_service.load_model_config(global_args.det_server_dir)
     ocr_service.init_det()
     if global_args.use_gpu:
         ocr_service.prepare_server(

deploy/pdserving/inference_to_serving.py

@@ -0,0 +1,32 @@
+# 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 os
+import argparse
+from paddle_serving_client.io import inference_model_to_serving
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model_dir", type=str)
+    parser.add_argument("--server_dir", type=str, default="serving_server_dir")
+    parser.add_argument("--client_dir", type=str, default="serving_client_dir")
+    return parser.parse_args()
+
+args = parse_args()
+inference_model_dir = args.model_dir
+serving_client_dir = os.path.join(args.model_dir, args.server_dir)
+serving_server_dir = os.path.join(args.model_dir, args.client_dir)
+feed_var_names, fetch_var_names = inference_model_to_serving(
+        inference_model_dir, serving_client_dir, serving_server_dir, model_filename="model", params_filename="params")
+
+print("success!")

deploy/pdserving/ocr_local_server.py

@@ -44,11 +44,11 @@ class TextSystemHelper(TextSystem):
         if self.use_angle_cls:
             self.clas_client = Debugger()
             self.clas_client.load_model_config(
-                global_args.cls_model_dir, gpu=True, profile=False)
+                global_args.cls_server_dir, gpu=True, profile=False)
             self.text_classifier = TextClassifierHelper(args)
         self.det_client = Debugger()
         self.det_client.load_model_config(
-            global_args.det_model_dir, gpu=True, profile=False)
+            global_args.det_server_dir, gpu=True, profile=False)
         self.fetch = ["save_infer_model/scale_0.tmp_0", "save_infer_model/scale_1.tmp_0"]
 
     def preprocess(self, img):
@@ -101,17 +101,20 @@ class OCRService(WebService):
             if ".lod" in x:
                 self.tmp_args[x] = fetch_map[x]
         rec_res = self.text_system.postprocess(outputs, self.tmp_args)
-        res = {
+        res = []
-            "pred_text": [x[0] for x in rec_res],
+        for i in range(len(rec_res)):
-            "score": [str(x[1]) for x in rec_res],
+            tmp_res = {
-            "pos": [x.tolist() for x in self.text_system.dt_boxes]
+                "text_region": self.text_system.dt_boxes[i].tolist(),
-        }
+                "text": rec_res[i][0],
+                "confidence": float(rec_res[i][1])
+            }
+            res.append(tmp_res)
         return res
 
 
 if __name__ == "__main__":
     ocr_service = OCRService(name="ocr")
-    ocr_service.load_model_config(global_args.rec_model_dir)
+    ocr_service.load_model_config(global_args.rec_server_dir)
     ocr_service.init_rec()
     if global_args.use_gpu:
         ocr_service.prepare_server(

deploy/pdserving/ocr_rpc_server.py

@@ -42,12 +42,14 @@ class TextSystemHelper(TextSystem):
         if self.use_angle_cls:
             self.clas_client = Client()
             self.clas_client.load_client_config(
-                "cls_infer_client/serving_client_conf.prototxt")
+                os.path.join(args.cls_client_dir, "serving_client_conf.prototxt")
+            )
             self.clas_client.connect(["127.0.0.1:9294"])
             self.text_classifier = TextClassifierHelper(args)
         self.det_client = Client()
         self.det_client.load_client_config(
-            "det_infer_client/serving_client_conf.prototxt")
+            os.path.join(args.det_client_dir, "serving_client_conf.prototxt")
+            )
         self.det_client.connect(["127.0.0.1:9293"])
         self.fetch = ["save_infer_model/scale_0.tmp_0", "save_infer_model/scale_1.tmp_0"]
 
@@ -101,17 +103,20 @@ class OCRService(WebService):
             if ".lod" in x:
                 self.tmp_args[x] = fetch_map[x]
         rec_res = self.text_system.postprocess(outputs, self.tmp_args)
-        res = {
+        res = []
-            "pred_text": [x[0] for x in rec_res],
+        for i in range(len(rec_res)):
-            "score": [str(x[1]) for x in rec_res],
+            tmp_res = {
-            "pos": [x.tolist() for x in self.text_system.dt_boxes]
+                "text_region": self.text_system.dt_boxes[i].tolist(),
-        }
+                "text": rec_res[i][0],
+                "confidence": float(rec_res[i][1])
+            }
+            res.append(tmp_res)
         return res
 
 
 if __name__ == "__main__":
     ocr_service = OCRService(name="ocr")
-    ocr_service.load_model_config(global_args.rec_model_dir)
+    ocr_service.load_model_config(global_args.rec_server_dir)
     ocr_service.init_rec()
     if global_args.use_gpu:
         ocr_service.prepare_server(

deploy/pdserving/ocr_web_client.py

@@ -1,40 +0,0 @@
-# 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.
-# -*- coding: utf-8 -*-
-
-import requests
-import json
-import cv2
-import base64
-import os, sys
-import time
-
-
-def cv2_to_base64(image):
-    #data = cv2.imencode('.jpg', image)[1]
-    return base64.b64encode(image).decode(
-        'utf8')  #data.tostring()).decode('utf8')
-
-
-headers = {"Content-type": "application/json"}
-url = "http://127.0.0.1:9292/ocr/prediction"
-test_img_dir = "../../doc/imgs/"
-for img_file in 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)
-    data = {"feed": [{"image": image}], "fetch": ["res"]}
-    r = requests.post(url=url, headers=headers, data=json.dumps(data))
-    rjson = r.json()
-    print(rjson)

deploy/pdserving/params.py

@@ -14,7 +14,8 @@ def read_params():
 
     #params for text detector
     cfg.det_algorithm = "DB"
-    cfg.det_model_dir = "./det_infer_server/"
+    cfg.det_server_dir = "../../ch_lite/ch_ppocr_mobile_v1.1_det_infer/serving_server_dir"
+    cfg.det_client_dir = "../../ch_lite/ch_ppocr_mobile_v1.1_det_infer/serving_client_dir"
     cfg.det_max_side_len = 960
 
     #DB parmas
@@ -29,19 +30,21 @@ def read_params():
 
     #params for text recognizer
     cfg.rec_algorithm = "CRNN"
-    cfg.rec_model_dir = "./rec_infer_server/"
+    cfg.rec_server_dir = "../../ch_lite/ch_ppocr_mobile_v1.1_rec_infer/serving_server_dir"
+    cfg.rec_client_dir = "../../ch_lite/ch_ppocr_mobile_v1.1_rec_infer/serving_client_dir"
 
     cfg.rec_image_shape = "3, 32, 320"
     cfg.rec_char_type = 'ch'
     cfg.rec_batch_num = 30
     cfg.max_text_length = 25
 
-    cfg.rec_char_dict_path = "./ppocr_keys_v1.txt"
+    cfg.rec_char_dict_path = "../../ppocr/utils/ppocr_keys_v1.txt"
     cfg.use_space_char = True
 
     #params for text classifier
     cfg.use_angle_cls = True
-    cfg.cls_model_dir = "./cls_infer_server/"
+    cfg.cls_server_dir = "../../ch_lite/ch_ppocr_mobile_v1.1_cls_infer/serving_server_dir"
+    cfg.cls_client_dir = "../../ch_lite/ch_ppocr_mobile_v1.1_cls_infer/serving_client_dir"
     cfg.cls_image_shape = "3, 48, 192"
     cfg.label_list = ['0', '180']
     cfg.cls_batch_num = 30

deploy/pdserving/pdserving_client.py

@@ -0,0 +1,117 @@
+# 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 os
+import sys
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, '..')))
+sys.path.append(os.path.abspath(os.path.join(__dir__, '../../')))
+
+from ppocr.utils.utility import initial_logger
+logger = initial_logger()
+import cv2
+import numpy as np
+import time
+from PIL import Image
+from ppocr.utils.utility import get_image_file_list
+from tools.infer.utility import draw_ocr, draw_boxes
+
+import requests
+import json
+import base64
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+
+def draw_server_result(image_file, res):
+    img = cv2.imread(image_file)
+    image = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
+    if len(res) == 0:
+        return np.array(image)
+    keys = res[0].keys()
+    if 'text_region' not in keys:  # for rec or clas, draw function is invalid 
+        logger.info("draw function is invalid for rec or clas!")
+        return None
+    elif 'text' not in keys:  # for ocr_det
+        logger.info("draw text boxes only!")
+        boxes = []
+        for dno in range(len(res)):
+            boxes.append(res[dno]['text_region'])
+        boxes = np.array(boxes)
+        draw_img = draw_boxes(image, boxes)
+        return draw_img
+    else:  # for ocr_system
+        logger.info("draw boxes and texts!")
+        boxes = []
+        texts = []
+        scores = []
+        for dno in range(len(res)):
+            boxes.append(res[dno]['text_region'])
+            texts.append(res[dno]['text'])
+            scores.append(res[dno]['confidence'])
+        boxes = np.array(boxes)
+        scores = np.array(scores)
+        draw_img = draw_ocr(image, boxes, texts, scores, drop_score=0.5, font_path="../../doc/simfang.ttf")
+        return draw_img
+
+
+def main(image_path):
+    image_file_list = get_image_file_list(image_path)
+    is_visualize = True
+    headers = {"Content-type": "application/json"}
+    url = "http://127.0.0.1:9292/ocr/prediction"
+    cnt = 0
+    total_time = 0
+    for image_file in image_file_list:
+        img = open(image_file, 'rb').read()
+        if img is None:
+            logger.info("error in loading image:{}".format(image_file))
+            continue
+
+        # 发送HTTP请求
+        starttime = time.time()
+        data = {"feed": [{"image": cv2_to_base64(img)}], "fetch": ["res"]}
+        r = requests.post(url=url, headers=headers, data=json.dumps(data))
+        elapse = time.time() - starttime
+        total_time += elapse
+        logger.info("Predict time of %s: %.3fs" % (image_file, elapse))
+        res = r.json()['result']
+        logger.info(res)
+
+        if is_visualize:
+            draw_img = draw_server_result(image_file, res)
+            if draw_img is not None:
+                draw_img_save = "./server_results/"
+                if not os.path.exists(draw_img_save):
+                    os.makedirs(draw_img_save)
+                cv2.imwrite(
+                    os.path.join(draw_img_save, os.path.basename(image_file)),
+                    draw_img[:, :, ::-1])
+                logger.info("The visualized image saved in {}".format(
+                    os.path.join(draw_img_save, os.path.basename(image_file))))
+        cnt += 1
+        if cnt % 100 == 0:
+            logger.info("{} processed".format(cnt))
+    logger.info("avg time cost: {}".format(float(total_time) / cnt))
+
+
+if __name__ == '__main__':
+    if len(sys.argv) != 2:
+        logger.info("Usage: %s image_path" % sys.argv[0])
+    else:
+        image_path = sys.argv[1]
+        main(image_path)

deploy/pdserving/readme.md

@@ -0,0 +1,195 @@
+[English](readme_en.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](../hubserving/readme.md)。		
+- 基于PaddleServing的部署：代码路径为"`./deploy/pdserving`"，按照本教程使用。	
+
+# Paddle Serving 服务部署	
+本教程将介绍基于[Paddle Serving](https://github.com/PaddlePaddle/Serving)部署PaddleOCR在线预测服务的详细步骤。
+- [快速启动服务](#快速启动服务)
+    - [1. 准备环境](#准备环境)
+    - [2. 转换模型](#转换模型)
+    - [3. 启动服务](#启动服务)
+- [发送预测请求](#发送预测请求)
+
+pdserving服务部署目录下包括`检测`、`识别`、`2阶段串联`三种服务部署工具，请根据需求选择相应的服务。目录结构如下：
+```
+deploy/pdserving/
+  └─  det_local_server.py     快速版 检测 服务端
+  └─  det_rpc_server.py       标准版 检测 服务端
+  └─  rec_local_server.py     快速版 识别 服务端
+  └─  rec_rpc_server.py       标准版 识别 服务端
+  └─  ocr_local_server.py     快速版 串联 服务端
+  └─  ocr_rpc_server.py       标准版 串联 服务端
+  └─  ocr_web_client.py       客户端
+  └─  params.py               配置文件
+```
+
+<a name="快速启动服务"></a>
+## 快速启动服务
+
+<a name="准备环境"></a>
+### 1. 准备环境
+环境版本要求：  
+- **CUDA版本：9.X/10.X**  
+- **CUDNN版本：7.X**  
+- **操作系统版本：Linux/Windows**  
+- **Python版本： 2.7/3.5/3.6/3.7** 
+
+**Python操作指南：**
+
+目前Serving用于OCR的部分功能还在测试当中，因此在这里我们给出[Servnig latest package](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md)
+大家根据自己的环境选择需要安装的whl包即可，例如以Python 3.6为例，执行下列命令：
+```
+# 安装服务端，CPU/GPU版本选择一个
+# GPU版本服务端
+# CUDA 9
+python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post9-py3-none-any.whl 
+# CUDA 10
+python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post10-py3-none-any.whl
+# CPU版本服务端
+python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server-0.0.0-py3-none-any.whl
+
+# 安装客户端和App包，CPU、GPU通用
+python -m pip install -U https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.0.0-cp35-none-any.whl https://paddle-serving.bj.bcebos.com/whl/paddle_serving_app-0.0.0-py3-none-any.whl
+
+# 安装其他依赖
+pip3.6 install func-timeout
+```
+
+<a name="转换模型"></a>
+## 2. 转换模型
+
+Paddle Serving无法直接用训练模型（checkpoints 模型）或推理模型（inference 模型）进行部署。Serving模型由两个文件夹构成，用于存放客户端和服务端的配置。本节介绍如何将推理模型转换为Paddle Serving可部署的模型。
+
+**以文本检测模型`ch_ppocr_mobile_v1.1_det_infer`为例，文本识别模型和方向分类器的转换同理。**
+
+首先下载推理模型：
+```shell
+wget -P ./inference/ https://paddleocr.bj.bcebos.com/20-09-22/mobile/det/ch_ppocr_mobile_v1.1_det_infer.tar && tar xf ./inference/ch_ppocr_mobile_v1.1_det_infer.tar -C ./inference/
+```
+然后运行如下python脚本进行转换，其中，使用参数`model_dir`指定待转换的推理模型路径：
+```
+python deploy/pdserving/inference_to_serving.py --model_dir ./inference/ch_ppocr_mobile_v1.1_det_infer
+```
+最终会在`ch_ppocr_mobile_v1.1_det_infer`目录下生成客户端和服务端的模型配置，结构如下：
+```
+/ch_ppocr_mobile_v1.1_det_infer/
+├── serving_client_dir # 客户端配置文件夹
+└── serving_server_dir # 服务端配置文件夹
+```
+
+<a name="启动服务"></a>
+## 3. 启动服务
+
+启动服务可以根据实际需求选择启动`标准版`或者`快速版`，两种方式的对比如下表：  
+
+|版本|特点|适用场景|
+|-|-|-|
+|标准版|稳定性高，分布式部署|适用于吞吐量大，需要跨机房部署的情况，只能用于Linux平台|
+|快速版|部署方便，预测速度快|适用于对预测速度要求高，迭代速度快的场景，可以支持Linux/Windows|
+
+
+**step 1. 配置环境变量**
+
+```
+# 以下两步的顺序不能反
+export PYTHONPATH=$PWD:$PYTHONPATH
+cd deploy/pdserving
+```
+
+**step 2. 修改配置参数**
+
+配置参数在`params.py`中，具体内容如下所示，可根据需要修改相关参数，如修改模型路径、修改后处理参数等。
+
+```
+def read_params():
+    cfg = Config()
+    #use gpu
+    cfg.use_gpu = False #是否使用GPU，False代表使用CPU
+    cfg.use_pdserving = True  #使用paddle serving部署时必须为True
+
+    #params for text detector
+    cfg.det_algorithm = "DB"
+    cfg.det_server_dir = "../../inference/ch_ppocr_mobile_v1.1_det_infer/serving_server_dir"
+    cfg.det_client_dir = "../../inference/ch_ppocr_mobile_v1.1_det_infer/serving_client_dir"
+    cfg.det_max_side_len = 960
+
+    #DB parmas
+    cfg.det_db_thresh =0.3
+    cfg.det_db_box_thresh =0.5
+    cfg.det_db_unclip_ratio =2.0
+
+    #EAST parmas
+    cfg.det_east_score_thresh = 0.8
+    cfg.det_east_cover_thresh = 0.1
+    cfg.det_east_nms_thresh = 0.2
+
+    #params for text recognizer
+    cfg.rec_algorithm = "CRNN"
+    cfg.rec_server_dir = "../../inference/ch_ppocr_mobile_v1.1_rec_infer/serving_server_dir"
+    cfg.rec_client_dir = "../../inference/ch_ppocr_mobile_v1.1_rec_infer/serving_client_dir"
+
+    cfg.rec_image_shape = "3, 32, 320"
+    cfg.rec_char_type = 'ch'
+    cfg.rec_batch_num = 30
+    cfg.max_text_length = 25
+
+    cfg.rec_char_dict_path = "../../ppocr/utils/ppocr_keys_v1.txt"
+    cfg.use_space_char = True
+
+    #params for text classifier
+    cfg.use_angle_cls = True
+    cfg.cls_server_dir = "../../inference/ch_ppocr_mobile_v1.1_cls_infer/serving_server_dir"
+    cfg.cls_client_dir = "../../inference/ch_ppocr_mobile_v1.1_cls_infer/serving_client_dir"
+    cfg.cls_image_shape = "3, 48, 192"
+    cfg.label_list = ['0', '180']
+    cfg.cls_batch_num = 30
+    cfg.cls_thresh = 0.9
+
+    return cfg
+```
+
+**step 3_1. 启动独立的检测服务或识别服务**
+
+如果只需要搭建检测服务或识别服务，一行命令即可，检测服务的启动方式如下，识别同理。检测+识别的串联服务请直接跳至step 3_2。
+
+```
+# 启动文本检测服务，标准版/快速版 二选一
+python det_rpc_server.py #标准版，Linux用户
+python det_local_server.py #快速版，Windows/Linux用户
+```
+
+**step 3_2. 启动文本检测、识别串联的服务**
+
+如果需要搭建检测+识别的串联服务，快速版与step 3_1中的独立服务启动方式相同，但标准版略有不同，具体步骤如下：
+
+```
+# 标准版，Linux用户
+# GPU用户
+# 启动检测服务
+python -m paddle_serving_server_gpu.serve --model inference/ch_ppocr_mobile_v1.1_det_infer/serving_server_dir/ --port 9293 --gpu_id 0
+# 启动方向分类器服务
+python -m paddle_serving_server_gpu.serve --model inference/ch_ppocr_mobile_v1.1_cls_infer/serving_server_dir/ --port 9294 --gpu_id 0
+# 启动串联服务
+python ocr_rpc_server.py 
+
+# CPU用户
+# 启动检测服务
+python -m paddle_serving_server.serve --model inference/ch_ppocr_mobile_v1.1_det_infer/serving_server_dir/ --port 9293
+# 启动方向分类器服务
+python -m paddle_serving_server.serve --model ch_ppocr_mobile_v1.1_cls_infer/serving_server_dir/ --port 9294
+# 启动串联服务
+python ocr_rpc_server.py
+
+# 快速版，Windows/Linux用户
+python ocr_local_server.py 
+```
+
+<a name="发送预测请求"></a>
+## 发送预测请求
+以上所有单独或串联的服务均可使用如下客户端进行访问：
+```
+python pdserving_client.py image_path
+```
+

deploy/pdserving/rec_local_server.py

@@ -153,16 +153,18 @@ class OCRService(WebService):
             if ".lod" in x:
                 self.tmp_args[x] = fetch_map[x]
         rec_res = self.text_recognizer.postprocess(outputs, self.tmp_args)
-        res = {
+        res = []
-            "pred_text": [x[0] for x in rec_res],
+        for i in range(len(rec_res)):
-            "score": [str(x[1]) for x in rec_res]
+            res.append({
-        }
+                "text": rec_res[i][0],
+                "confidence": float(rec_res[i][1])
+            })
         return res
 
 
 if __name__ == "__main__":
     ocr_service = OCRService(name="ocr")
-    ocr_service.load_model_config(global_args.rec_model_dir)
+    ocr_service.load_model_config(global_args.rec_server_dir)
     ocr_service.init_rec()
     if global_args.use_gpu:
         ocr_service.prepare_server(

deploy/pdserving/rec_rpc_server.py

@@ -158,16 +158,18 @@ class OCRService(WebService):
             if ".lod" in x:
                 self.tmp_args[x] = fetch_map[x]
         rec_res = self.text_recognizer.postprocess(outputs, self.tmp_args)
-        res = {
+        res = []
-            "pred_text": [x[0] for x in rec_res],
+        for i in range(len(rec_res)):
-            "score": [str(x[1]) for x in rec_res]
+            res.append({
-        }
+                "text": rec_res[i][0],
+                "confidence": float(rec_res[i][1])
+            })
         return res
 
 
 if __name__ == "__main__":
     ocr_service = OCRService(name="ocr")
-    ocr_service.load_model_config(global_args.rec_model_dir)
+    ocr_service.load_model_config(global_args.rec_server_dir)
     ocr_service.init_rec()
     if global_args.use_gpu:
         ocr_service.prepare_server(