add paddle serving doc

deploy/README.md

@@ -10,3 +10,4 @@
 
 [4. 移动端部署(仅支持Android)](./lite)
 
+[5. 使用PaddleServing部署](./paddle-serving)

deploy/paddle-serving/README.md

+# 通过PaddleServing部署服务
+
+## 1.简介
+
+PaddleServing是Paddle的在线预测服务框架，可以快速部署训练好的模型用于在线预测。更多信息请参考[PaddleServing 主页](https://github.com/PaddlePaddle/Serving)。本文中将通过unet模型示例，展示预测服务的部署和预测过程。
+
+## 2.安装Paddle Serving
+
+目前PaddleServing的正式版本为0.3.2版本，本文中的示例需要develop版本的paddle_serving_app，请从[链接](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md#app)中下载并安装。
+
+服务端安装：
+
+```shell
+pip install paddle_serving_server==0.3.2 #CPU
+pip install paddle_serving_server_gpu==0.3.2.post9 #GPU with CUDA9.0
+pip install paddle_serving_server_gpu==0.3.2.post9 #GPU with CUDA10.0
+```
+
+客户端安装：
+
+```shell
+pip install paddle_serving_client==0.3.2
+```
+
+## 3.导出预测模型
+
+
+通过训练得到一个满足要求的模型后，如果想要将该模型接入到PaddleServing服务，我们需要通过[`pdseg/export_serving_model.py`](../../pdseg/export_serving_model.py)来导出该模型。
+
+该脚本的使用方法和`train.py/eval.py/vis.py`完全一样。
+
+### FLAGS
+
+|FLAG|用途|默认值|备注|
+|-|-|-|-|
+|--cfg|配置文件路径|None||
+
+### 使用示例
+
+我们使用[训练/评估/可视化](./usage.md)一节中训练得到的模型进行试用，命令如下
+
+```shell
+python pdseg/export_serving_model.py --cfg configs/unet_optic.yaml TEST.TEST_MODEL ./saved_model/unet_optic/final
+```
+
+预测模型会导出到`freeze_model`目录，包括`serving_server`和`serving_client`两个子目录。
+
+`freeze_model/serving_server`目录下包含了模型文件和serving server端配置文件，`freeze_model/serving_client`目录下包含了serving client端配置文件。
+
+## 4.部署预测服务
+
+```shell
+python -m paddle_serving_server.serve --model unet_model/ --port 9494 # CPU
+python -m paddle_serving_server_gpu.serve --model unet_model --port 9494 --gpu_ids 0 #GPU
+```
+
+## 5.执行预测
+
+```python
+#seg_client.py
+from paddle_serving_client import Client
+from paddle_serving_app.reader import Sequential, File2Image, Resize, Transpose, BGR2RGB, SegPostprocess, Normalize, Div
+import sys
+import cv2
+
+client = Client()
+client.load_client_config("unet_client/serving_client_conf.prototxt")
+client.connect(["127.0.0.1:9494"])
+
+preprocess = Sequential([
+    File2Image(), Resize(
+        (512, 512), interpolation=cv2.INTER_LINEAR), Div(255.0),
+    Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5], False), Transpose((2, 0, 1))
+])
+
+postprocess = SegPostprocess(2)
+
+filename = "N0060.jpg"
+im = preprocess(filename)
+fetch_map = client.predict(feed={"image": im}, fetch=["transpose_1.tmp_0"])
+fetch_map["filename"] = filename
+postprocess(fetch_map)
+```
+
+脚本执行之后，当前目录下生成处理后的图片。
+
+完整的部署示例请参考PaddleServing的[unet示例](https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/unet_for_image_seg)。

pdseg/export_serving_model.py

+# coding: utf8
+# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# 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
+import time
+import pprint
+import cv2
+import argparse
+import numpy as np
+import paddle.fluid as fluid
+
+from utils.config import cfg
+from models.model_builder import build_model
+from models.model_builder import ModelPhase
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='PaddleSeg Inference Model Exporter')
+    parser.add_argument(
+        '--cfg',
+        dest='cfg_file',
+        help='Config file for training (and optionally testing)',
+        default=None,
+        type=str)
+    parser.add_argument('opts',
+                        help='See utils/config.py for all options',
+                        default=None,
+                        nargs=argparse.REMAINDER)
+    if len(sys.argv) == 1:
+        parser.print_help()
+        sys.exit(1)
+    return parser.parse_args()
+
+
+def export_inference_config():
+    deploy_cfg = '''DEPLOY:
+        USE_GPU : 1
+        USE_PR : 0
+        MODEL_PATH : "%s"
+        MODEL_FILENAME : "%s"
+        PARAMS_FILENAME : "%s"
+        EVAL_CROP_SIZE : %s
+        MEAN : %s
+        STD : %s
+        IMAGE_TYPE : "%s"
+        NUM_CLASSES : %d
+        CHANNELS : %d
+        PRE_PROCESSOR : "SegPreProcessor"
+        PREDICTOR_MODE : "ANALYSIS"
+        BATCH_SIZE : 1
+    ''' % (cfg.FREEZE.SAVE_DIR, cfg.FREEZE.MODEL_FILENAME,
+           cfg.FREEZE.PARAMS_FILENAME, cfg.EVAL_CROP_SIZE, cfg.MEAN, cfg.STD,
+           cfg.DATASET.IMAGE_TYPE, cfg.DATASET.NUM_CLASSES, len(cfg.STD))
+    if not os.path.exists(cfg.FREEZE.SAVE_DIR):
+        os.mkdir(cfg.FREEZE.SAVE_DIR)
+    yaml_path = os.path.join(cfg.FREEZE.SAVE_DIR, 'deploy.yaml')
+    with open(yaml_path, "w") as fp:
+        fp.write(deploy_cfg)
+    return yaml_path
+
+
+def export_serving_model(args):
+    """
+    Export PaddlePaddle inference model for prediction depolyment and serving.
+    """
+    print("Exporting serving model...")
+    startup_prog = fluid.Program()
+    infer_prog = fluid.Program()
+    image, logit_out = build_model(infer_prog,
+                                   startup_prog,
+                                   phase=ModelPhase.PREDICT)
+
+    # Use CPU for exporting inference model instead of GPU
+    place = fluid.CPUPlace()
+    exe = fluid.Executor(place)
+    exe.run(startup_prog)
+    infer_prog = infer_prog.clone(for_test=True)
+
+    if os.path.exists(cfg.TEST.TEST_MODEL):
+        print('load test model:', cfg.TEST.TEST_MODEL)
+        try:
+            fluid.load(infer_prog, os.path.join(cfg.TEST.TEST_MODEL, 'model'),
+                       exe)
+        except:
+            fluid.io.load_params(exe,
+                                 cfg.TEST.TEST_MODEL,
+                                 main_program=infer_prog)
+    else:
+        print("TEST.TEST_MODEL diretory is empty!")
+        exit(-1)
+
+    from paddle_serving_client.io import save_model
+    save_model(
+        cfg.FREEZE.SAVE_DIR + "/serving_server",
+        cfg.FREEZE.SAVE_DIR + "/serving_client",
+        {image.name: image},
+        {logit_out.name: logit_out},
+        infer_prog,
+    )
+    print("Serving model exported!")
+    print("Exporting serving model config...")
+    deploy_cfg_path = export_inference_config()
+    print("Serving model saved : [%s]" % (deploy_cfg_path))
+
+
+def main():
+    args = parse_args()
+    if args.cfg_file is not None:
+        cfg.update_from_file(args.cfg_file)
+    if args.opts:
+        cfg.update_from_list(args.opts)
+    cfg.check_and_infer()
+    print(pprint.pformat(cfg))
+    export_serving_model(args)
+
+
+if __name__ == '__main__':
+    main()