Merge branch 'develop' of https://github.com/cuicheng01/PaddleClas into develop

deploy/cpp/CMakeLists.txt

 project(clas_system CXX C)
+cmake_minimum_required(VERSION 3.14)
 
 option(WITH_MKL        "Compile demo with MKL/OpenBlas support, default use MKL."       ON)
 option(WITH_GPU        "Compile demo with GPU/CPU, default use CPU."                    OFF)
@@ -13,7 +14,6 @@ SET(TENSORRT_DIR "" CACHE PATH "Compile demo with TensorRT")
 
 set(DEMO_NAME "clas_system")
 
-
 macro(safe_set_static_flag)
     foreach(flag_var
         CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
@@ -198,6 +198,10 @@ endif()
 
 set(DEPS ${DEPS} ${OpenCV_LIBS})
 
+include(FetchContent)
+include(external-cmake/auto-log.cmake)
+include_directories(${FETCHCONTENT_BASE_DIR}/extern_autolog-src)
+
 AUX_SOURCE_DIRECTORY(./src SRCS)
 add_executable(${DEMO_NAME} ${SRCS})
 

deploy/cpp/external-cmake/auto-log.cmake

+find_package(Git REQUIRED)
+include(FetchContent)
+
+set(FETCHCONTENT_BASE_DIR "${CMAKE_CURRENT_BINARY_DIR}/third-party")
+
+FetchContent_Declare(
+  extern_Autolog
+  PREFIX autolog
+  GIT_REPOSITORY https://github.com/LDOUBLEV/AutoLog.git
+  GIT_TAG        main
+)
+FetchContent_MakeAvailable(extern_Autolog)

deploy/cpp/include/cls.h

@@ -61,7 +61,7 @@ public:
   void LoadModel(const std::string &model_path, const std::string &params_path);
 
   // Run predictor
-  double Run(cv::Mat &img);
+  double Run(cv::Mat &img, std::vector<double> *times);
 
 private:
   std::shared_ptr<Predictor> predictor_;

deploy/cpp/include/cls_config.h

-// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+// 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.
@@ -36,8 +36,7 @@ public:
 
     this->gpu_mem = stoi(config_map_["gpu_mem"]);
 
-    this->cpu_math_library_num_threads =
-        stoi(config_map_["cpu_math_library_num_threads"]);
+    this->cpu_threads = stoi(config_map_["cpu_threads"]);
 
     this->use_mkldnn = bool(stoi(config_map_["use_mkldnn"]));
 
@@ -51,6 +50,8 @@ public:
     this->resize_short_size = stoi(config_map_["resize_short_size"]);
 
     this->crop_size = stoi(config_map_["crop_size"]);
+
+    this->benchmark = bool(stoi(config_map_["benchmark"]));
   }
 
   bool use_gpu = false;
@@ -59,12 +60,13 @@ public:
 
   int gpu_mem = 4000;
 
-  int cpu_math_library_num_threads = 1;
+  int cpu_threads = 1;
 
   bool use_mkldnn = false;
 
   bool use_tensorrt = false;
   bool use_fp16 = false;
+  bool benchmark = false;
 
   std::string cls_model_path;
 

deploy/cpp/src/cls.cpp

@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include <chrono>
 #include <include/cls.h>
 
 namespace PaddleClas {
@@ -53,11 +52,12 @@ void Classifier::LoadModel(const std::string &model_path,
   this->predictor_ = CreatePredictor(config);
 }
 
-double Classifier::Run(cv::Mat &img) {
+double Classifier::Run(cv::Mat &img, std::vector<double> *times) {
   cv::Mat srcimg;
   cv::Mat resize_img;
   img.copyTo(srcimg);
 
+  auto preprocess_start = std::chrono::system_clock::now();
   this->resize_op_.Run(img, resize_img, this->resize_short_size_);
 
   this->crop_op_.Run(resize_img, this->crop_size_);
@@ -70,7 +70,9 @@ double Classifier::Run(cv::Mat &img) {
   auto input_names = this->predictor_->GetInputNames();
   auto input_t = this->predictor_->GetInputHandle(input_names[0]);
   input_t->Reshape({1, 3, resize_img.rows, resize_img.cols});
-  auto start = std::chrono::system_clock::now();
+  auto preprocess_end = std::chrono::system_clock::now();
+
+  auto infer_start = std::chrono::system_clock::now();
   input_t->CopyFromCpu(input.data());
   this->predictor_->Run();
 
@@ -83,21 +85,29 @@ double Classifier::Run(cv::Mat &img) {
 
   out_data.resize(out_num);
   output_t->CopyToCpu(out_data.data());
-  auto end = std::chrono::system_clock::now();
-  auto duration =
-      std::chrono::duration_cast<std::chrono::microseconds>(end - start);
-  double cost_time = double(duration.count()) *
-                     std::chrono::microseconds::period::num /
-                     std::chrono::microseconds::period::den;
+  auto infer_end = std::chrono::system_clock::now();
 
+  auto postprocess_start = std::chrono::system_clock::now();
   int maxPosition =
       max_element(out_data.begin(), out_data.end()) - out_data.begin();
+  auto postprocess_end = std::chrono::system_clock::now();
+
+  std::chrono::duration<float> preprocess_diff =
+      preprocess_end - preprocess_start;
+  times->push_back(double(preprocess_diff.count() * 1000));
+  std::chrono::duration<float> inference_diff = infer_end - infer_start;
+  double inference_cost_time = double(inference_diff.count() * 1000);
+  times->push_back(inference_cost_time);
+  std::chrono::duration<float> postprocess_diff =
+      postprocess_end - postprocess_start;
+  times->push_back(double(postprocess_diff.count() * 1000));
+
   std::cout << "result: " << std::endl;
   std::cout << "\tclass id: " << maxPosition << std::endl;
   std::cout << std::fixed << std::setprecision(10)
             << "\tscore: " << double(out_data[maxPosition]) << std::endl;
 
-  return cost_time;
+  return inference_cost_time;
 }
 
 } // namespace PaddleClas

deploy/cpp/src/main.cpp

-// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+// 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.
@@ -26,6 +26,7 @@
 #include <fstream>
 #include <numeric>
 
+#include <auto_log/autolog.h>
 #include <include/cls.h>
 #include <include/cls_config.h>
 
@@ -61,11 +62,12 @@ int main(int argc, char **argv) {
 
   Classifier classifier(config.cls_model_path, config.cls_params_path,
                         config.use_gpu, config.gpu_id, config.gpu_mem,
-                        config.cpu_math_library_num_threads, config.use_mkldnn,
+                        config.cpu_threads, config.use_mkldnn,
                         config.use_tensorrt, config.use_fp16,
                         config.resize_short_size, config.crop_size);
 
   double elapsed_time = 0.0;
+  std::vector<double> cls_times;
   int warmup_iter = img_files_list.size() > 5 ? 5 : 0;
   for (int idx = 0; idx < img_files_list.size(); ++idx) {
     std::string img_path = img_files_list[idx];
@@ -78,7 +80,7 @@ int main(int argc, char **argv) {
 
     cv::cvtColor(srcimg, srcimg, cv::COLOR_BGR2RGB);
 
-    double run_time = classifier.Run(srcimg);
+    double run_time = classifier.Run(srcimg, &cls_times);
     if (idx >= warmup_iter) {
       elapsed_time += run_time;
       std::cout << "Current image path: " << img_path << std::endl;
@@ -90,5 +92,16 @@ int main(int argc, char **argv) {
     }
   }
 
+  std::string presion = "fp32";
+
+  if (config.use_fp16)
+    presion = "fp16";
+  if (config.benchmark) {
+    AutoLogger autolog("Classification", config.use_gpu, config.use_tensorrt,
+                       config.use_mkldnn, config.cpu_threads, 1,
+                       "1, 3, 224, 224", presion, cls_times,
+                       img_files_list.size());
+    autolog.report();
+  }
   return 0;
 }

deploy/cpp/tools/build.sh

-OPENCV_DIR=/PaddleClas/opencv-3.4.7/opencv3/
-LIB_DIR=/PaddleClas/fluid_inference/
+OPENCV_DIR=/work/project/project/cpp_infer/opencv-3.4.7/opencv3
+LIB_DIR=/work/project/project/cpp_infer/paddle_inference/
 CUDA_LIB_DIR=/usr/local/cuda/lib64
 CUDNN_LIB_DIR=/usr/lib/x86_64-linux-gnu/
 

deploy/cpp/tools/config.txt

@@ -2,7 +2,7 @@
 use_gpu  0
 gpu_id  0
 gpu_mem  4000
-cpu_math_library_num_threads  10
+cpu_threads  10
 use_mkldnn 1
 use_tensorrt 0
 use_fp16 0
@@ -12,3 +12,6 @@ cls_model_path  /PaddleClas/inference/cls_infer.pdmodel
 cls_params_path /PaddleClas/inference/cls_infer.pdiparams
 resize_short_size 256
 crop_size 224
+
+# for log env info
+benchmark 0

deploy/hubserving/clas/module.py

@@ -76,8 +76,7 @@ class ClasSystem(nn.Layer):
         starttime = time.time()
         outputs = self.cls_predictor.predict(inputs)
         elapse = time.time() - starttime
-        preds = self.cls_predictor.postprocess(outputs)
-        return {"prediction": preds, "elapse": elapse}
+        return {"prediction": outputs, "elapse": elapse}
 
     @serving
     def serving_method(self, images, revert_params):

deploy/paddleserving/README.md

+# PaddleClas Pipeline WebService
+
+(English|[简体中文](./README_CN.md))
+
+PaddleClas 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/paddleserving`". 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 ResNet50_vd 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
+
+PaddleClas operating environment and PaddleServing operating environment are needed.
+
+1. Please prepare PaddleClas operating environment reference [link](../../docs/zh_CN/tutorials/install.md).
+   Download the corresponding paddle whl package according to the environment, it is recommended to install version 2.1.0.
+
+2. The steps of PaddleServing operating environment prepare are as follows:
+
+    Install serving which used to start the service
+    ```
+    pip3 install paddle-serving-server==0.6.1 # for CPU
+    pip3 install paddle-serving-server-gpu==0.6.1 # for GPU
+    # Other GPU environments need to confirm the environment and then choose to execute the following commands
+    pip3 install paddle-serving-server-gpu==0.6.1.post101 # GPU with CUDA10.1 + TensorRT6
+    pip3 install paddle-serving-server-gpu==0.6.1.post11 # GPU with CUDA11 + TensorRT7
+    ```
+
+3. Install the client to send requests to the service
+    In [download link](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md) find the client installation package corresponding to the python version.
+    The python3.7 version is recommended here:
+
+    ```
+    wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_client-0.0.0-cp37-none-any.whl
+    pip3 install paddle_serving_client-0.0.0-cp37-none-any.whl
+    ```
+
+4. Install serving-app
+    ```
+    pip3 install paddle-serving-app==0.6.1
+    ```
+
+   **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 of ResNet50_vd
+```
+# Download and unzip the ResNet50_vd model
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/ResNet50_vd_infer.tar  && tar xf ResNet50_vd_infer.tar
+```
+
+Then, you can use installed paddle_serving_client tool to convert inference model to mobile model.
+```
+#  ResNet50_vd model conversion
+python3 -m paddle_serving_client.convert --dirname ./ResNet50_vd_infer/ \
+                                         --model_filename inference.pdmodel  \
+                                         --params_filename inference.pdiparams \
+                                         --serving_server ./ResNet50_vd_serving/ \
+                                         --serving_client ./ResNet50_vd_client/
+```
+
+After the ResNet50_vd inference model is converted, there will be additional folders of `ResNet50_vd_serving` and `ResNet50_vd_client` in the current folder, with the following format:
+```
+|- ResNet50_vd_client/
+  |- __model__  
+  |- __params__
+  |- serving_server_conf.prototxt  
+  |- serving_server_conf.stream.prototxt
+
+|- ResNet50_vd_client
+  |- serving_client_conf.prototxt  
+  |- serving_client_conf.stream.prototxt
+```
+
+Once you have the model file for deployment, you need to change the alias name in `serving_server_conf.prototxt`: Change `alias_name` in `feed_var` to `image`, change `alias_name` in `fetch_var` to `prediction`,
+The modified serving_server_conf.prototxt file is as follows:
+```
+feed_var {
+  name: "inputs"
+  alias_name: "image"
+  is_lod_tensor: false
+  feed_type: 1
+  shape: 3
+  shape: 224
+  shape: 224
+}
+fetch_var {
+  name: "save_infer_model/scale_0.tmp_1"
+  alias_name: "prediction"
+  is_lod_tensor: true
+  fetch_type: 1
+  shape: -1
+}
+```
+
+<a name="paddle-serving-pipeline-deployment"></a>
+## Paddle Serving pipeline deployment
+
+1. Download the PaddleClas code, if you have already downloaded it, you can skip this step.
+    ```
+    git clone https://github.com/PaddlePaddle/PaddleClas
+
+    # Enter the working directory  
+    cd PaddleClas/deploy/paddleserving/
+    ```
+
+    The paddleserving directory contains the code to start the pipeline service and send prediction requests, including:
+    ```
+    __init__.py
+    config.yml                # configuration file of starting the service
+    pipeline_http_client.py   # script to send pipeline prediction request by http
+    pipeline_rpc_client.py    # script to send pipeline prediction request by rpc
+    resnet50_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 classification_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)  
+
+    Adjust the number of concurrency in config.yml to get the largest QPS. 
+
+    ```
+    op:
+        concurrency: 8
+        ...
+    ```
+
+    Multiple service requests can be sent at the same time if necessary.
+
+    The predicted performance data will be automatically written into the `PipelineServingLogs/pipeline.tracer` file.
+
+<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/paddleserving/README_CN.md

+# PaddleClas 服务化部署
+
+([English](./README.md)|简体中文)
+
+PaddleClas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/paddleserving`"，按照本教程使用。
+
+# 基于PaddleServing的服务部署
+
+本文档以经典的ResNet50_vd模型为例，介绍如何使用[PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)工具部署PaddleClas
+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PaddleClas的运行环境和PaddleServing的运行环境。
+
+- 准备PaddleClas的[运行环境](../../docs/zh_CN/tutorials/install.md), 根据环境下载对应的paddle whl包，推荐安装2.1.0版本
+
+- 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==0.6.1 # for CPU
+    pip3 install paddle-serving-server-gpu==0.6.1 # 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==0.6.1.post101 # GPU with CUDA10.1 + TensorRT6
+    pip3 install paddle-serving-server-gpu==0.6.1.post11 # GPU with CUDA11 + TensorRT7
+    ```
+
+2. 安装client，用于向服务发送请求
+    在[下载链接](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md)中找到对应python版本的client安装包，这里推荐python3.7版本：
+
+    ```
+    wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_client-0.0.0-cp37-none-any.whl
+    pip3 install paddle_serving_client-0.0.0-cp37-none-any.whl
+    ```
+
+3. 安装serving-app
+    ```
+    pip3 install paddle-serving-app==0.6.1
+    ```
+    **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ResNet50_vd的inference模型
+```
+# 下载并解压ResNet50_vd模型
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/ResNet50_vd_infer.tar && tar xf ResNet50_vd_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ResNet50_vd模型
+python3 -m paddle_serving_client.convert --dirname ./ResNet50_vd_infer/ \
+                                         --model_filename inference.pdmodel  \
+                                         --params_filename inference.pdiparams \
+                                         --serving_server ./ResNet50_vd_serving/ \
+                                         --serving_client ./ResNet50_vd_client/
+```
+ResNet50_vd推理模型转换完成后，会在当前文件夹多出`ResNet50_vd_serving` 和`ResNet50_vd_client`的文件夹，具备如下格式：
+```
+|- ResNet50_vd_client/
+  |- __model__  
+  |- __params__
+  |- serving_server_conf.prototxt  
+  |- serving_server_conf.stream.prototxt
+
+|- ResNet50_vd_client
+  |- serving_client_conf.prototxt  
+  |- serving_client_conf.stream.prototxt
+
+```
+得到模型文件之后，需要修改serving_server_conf.prototxt中的alias名字： 将`feed_var`中的`alias_name`改为`image`, 将`fetch_var`中的`alias_name`改为`prediction`, 
+修改后的serving_server_conf.prototxt内容如下：
+```
+feed_var {
+  name: "inputs"
+  alias_name: "image"
+  is_lod_tensor: false
+  feed_type: 1
+  shape: 3
+  shape: 224
+  shape: 224
+}
+fetch_var {
+  name: "save_infer_model/scale_0.tmp_1"
+  alias_name: "prediction"
+  is_lod_tensor: true
+  fetch_type: 1
+  shape: -1
+}
+```
+
+<a name="部署"></a>
+## Paddle Serving pipeline部署
+
+1. 下载PaddleClas代码，若已下载可跳过此步骤
+    ```
+    git clone https://github.com/PaddlePaddle/PaddleClas
+
+    # 进入到工作目录
+    cd PaddleClas/deploy/paddleserving/
+    ```
+    paddleserving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
+    pipeline_http_client.py    # http方式发送pipeline预测请求的脚本
+    pipeline_rpc_client.py     # rpc方式发送pipeline预测请求的脚本
+    resnet50_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 classification_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)
+
+    调整 config.yml 中的并发个数可以获得最大的QPS
+    ```
+    op:
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 8
+        ...
+    ```
+    有需要的话可以同时发送多个服务请求
+
+    预测性能数据会被自动写入 `PipelineServingLogs/pipeline.tracer` 文件中。
+
+<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/paddleserving/classification_web_service.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 sys
+from paddle_serving_app.reader import Sequential, URL2Image, Resize, CenterCrop, RGB2BGR, Transpose, Div, Normalize, Base64ToImage
+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 base64, cv2
+
+class ImagenetOp(Op):
+    def init_op(self):
+        self.seq = Sequential([
+            Resize(256), CenterCrop(224), RGB2BGR(), Transpose((2, 0, 1)),
+            Div(255), Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225],
+                                True)
+        ])
+        self.label_dict = {}
+        label_idx = 0
+        with open("imagenet.label") as fin:
+            for line in fin:
+                self.label_dict[label_idx] = line.strip()
+                label_idx += 1
+
+    def preprocess(self, input_dicts, data_id, log_id):
+        (_, input_dict), = input_dicts.items()
+        batch_size = len(input_dict.keys())
+        imgs = []
+        for key in input_dict.keys():
+            data = base64.b64decode(input_dict[key].encode('utf8'))
+            data = np.fromstring(data, np.uint8)
+            im = cv2.imdecode(data, cv2.IMREAD_COLOR)
+            img = self.seq(im)
+            imgs.append(img[np.newaxis, :].copy())
+        input_imgs = np.concatenate(imgs, axis=0)
+        return {"image": input_imgs}, False, None, ""
+
+    def postprocess(self, input_dicts, fetch_dict, log_id):
+        score_list = fetch_dict["prediction"]
+        result = {"label": [], "prob": []}
+        for score in score_list:
+            score = score.tolist()
+            max_score = max(score)
+            result["label"].append(self.label_dict[score.index(max_score)]
+                                   .strip().replace(",", ""))
+            result["prob"].append(max_score)
+        result["label"] = str(result["label"])
+        result["prob"] = str(result["prob"])
+        return result, None, ""
+
+
+class ImageService(WebService):
+    def get_pipeline_response(self, read_op):
+        image_op = ImagenetOp(name="imagenet", input_ops=[read_op])
+        return image_op
+
+
+uci_service = ImageService(name="imagenet")
+uci_service.prepare_pipeline_config("config.yml")
+uci_service.run_service()

deploy/paddleserving/config.yml

+#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: 1
+
+#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: 18080
+rpc_port: 9993
+
+dag:
+    #op资源类型, True, 为线程模型；False，为进程模型
+    is_thread_op: False
+op:
+    imagenet:
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 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:
+
+            #uci模型路径
+            model_config: ResNet50_vd_serving
+
+            #计算硬件类型: 空缺时由devices决定(CPU/GPU)，0=cpu, 1=gpu, 2=tensorRT, 3=arm cpu, 4=kunlun xpu
+            device_type: 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: "0" # "0,1"
+
+            #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
+
+            #Fetch结果列表，以client_config中fetch_var的alias_name为准
+            fetch_list: ["prediction"] 

deploy/paddleserving/cpu_utilization.py

+import psutil
+cpu_utilization=psutil.cpu_percent(1,False)
+print('CPU_UTILIZATION:', cpu_utilization)
+

deploy/paddleserving/image_service_cpu.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 sys
-import base64
-from paddle_serving_server.web_service import WebService
-import utils
-
-
-class ImageService(WebService):
-    def __init__(self, name):
-        super(ImageService, self).__init__(name=name)
-        self.operators = self.create_operators()
-
-    def create_operators(self):
-        size = 224
-        img_mean = [0.485, 0.456, 0.406]
-        img_std = [0.229, 0.224, 0.225]
-        img_scale = 1.0 / 255.0
-        decode_op = utils.DecodeImage()
-        resize_op = utils.ResizeImage(resize_short=256)
-        crop_op = utils.CropImage(size=(size, size))
-        normalize_op = utils.NormalizeImage(
-            scale=img_scale, mean=img_mean, std=img_std)
-        totensor_op = utils.ToTensor()
-        return [decode_op, resize_op, crop_op, normalize_op, totensor_op]
-
-    def _process_image(self, data, ops):
-        for op in ops:
-            data = op(data)
-        return data
-
-    def preprocess(self, feed={}, fetch=[]):
-        feed_batch = []
-        for ins in feed:
-            if "image" not in ins:
-                raise ("feed data error!")
-            sample = base64.b64decode(ins["image"])
-            img = self._process_image(sample, self.operators)
-            feed_batch.append({"image": img})
-        return feed_batch, fetch
-
-
-image_service = ImageService(name="image")
-image_service.load_model_config(sys.argv[1])
-image_service.prepare_server(
-    workdir=sys.argv[2], port=int(sys.argv[3]), device="cpu")
-image_service.run_server()
-image_service.run_flask()

deploy/paddleserving/image_service_gpu.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 sys
-import base64
-from paddle_serving_server_gpu.web_service import WebService
-
-import utils
-
-
-class ImageService(WebService):
-    def __init__(self, name):
-        super(ImageService, self).__init__(name=name)
-        self.operators = self.create_operators()
-
-    def create_operators(self):
-        size = 224
-        img_mean = [0.485, 0.456, 0.406]
-        img_std = [0.229, 0.224, 0.225]
-        img_scale = 1.0 / 255.0
-        decode_op = utils.DecodeImage()
-        resize_op = utils.ResizeImage(resize_short=256)
-        crop_op = utils.CropImage(size=(size, size))
-        normalize_op = utils.NormalizeImage(
-            scale=img_scale, mean=img_mean, std=img_std)
-        totensor_op = utils.ToTensor()
-        return [decode_op, resize_op, crop_op, normalize_op, totensor_op]
-
-    def _process_image(self, data, ops):
-        for op in ops:
-            data = op(data)
-        return data
-
-    def preprocess(self, feed={}, fetch=[]):
-        feed_batch = []
-        for ins in feed:
-            if "image" not in ins:
-                raise ("feed data error!")
-            sample = base64.b64decode(ins["image"])
-            img = self._process_image(sample, self.operators)
-            feed_batch.append({"image": img})
-        return feed_batch, fetch
-
-
-image_service = ImageService(name="image")
-image_service.load_model_config(sys.argv[1])
-image_service.set_gpus("0")
-image_service.prepare_server(
-    workdir=sys.argv[2], port=int(sys.argv[3]), device="gpu")
-image_service.run_server()
-image_service.run_flask()

deploy/paddleserving/pipeline_http_client.py

+import requests
+import json
+import base64
+import os
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+if __name__ == "__main__":
+    url = "http://127.0.0.1:18080/imagenet/prediction"
+    with open(os.path.join(".", "daisy.jpg"), 'rb') as file:
+        image_data1 = file.read()
+    image = cv2_to_base64(image_data1)
+    data = {"key": ["image"], "value": [image]}
+    for i in range(100):
+        r = requests.post(url=url, data=json.dumps(data))
+        print(r.json())

deploy/paddleserving/image_http_client.py → deploy/paddleserving/pipeline_rpc_client.py

@@ -11,36 +11,23 @@
 # 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 requests
+try:
+    from paddle_serving_server_gpu.pipeline import PipelineClient
+except ImportError:
+    from paddle_serving_server.pipeline import PipelineClient
 import base64
-import json
-import sys
-import numpy as np
-
-py_version = sys.version_info[0]
-
 
-def predict(image_path, server):
-
-    with open(image_path, "rb") as f:
-        image = base64.b64encode(f.read()).decode("utf-8")
-    req = json.dumps({"feed": [{"image": image}], "fetch": ["prediction"]})
-    r = requests.post(
-        server, data=req, headers={"Content-Type": "application/json"})
-    try:
-        pred = r.json()["result"]["prediction"][0]
-        cls_id = np.argmax(pred)
-        score = pred[cls_id]
-        pred = {"cls_id": cls_id, "score": score}
-        return pred
-    except ValueError:
-        print(r.text)
-    return r
+client = PipelineClient()
+client.connect(['127.0.0.1:9993'])
 
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
 
 if __name__ == "__main__":
-    server = "http://127.0.0.1:{}/image/prediction".format(sys.argv[1])
-    image_file = sys.argv[2]
-    res = predict(image_file, server)
-    print("res:", res)
+    with open("daisy.jpg", '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=["label", "prob"])
+        print(ret)

deploy/paddleserving/utils.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 cv2
-import numpy as np
-
-
-class DecodeImage(object):
-    def __init__(self, to_rgb=True):
-        self.to_rgb = to_rgb
-
-    def __call__(self, img):
-        data = np.frombuffer(img, dtype='uint8')
-        img = cv2.imdecode(data, 1)
-        if self.to_rgb:
-            assert img.shape[2] == 3, 'invalid shape of image[%s]' % (
-                img.shape)
-            img = img[:, :, ::-1]
-
-        return img
-
-
-class ResizeImage(object):
-    def __init__(self, resize_short=None):
-        self.resize_short = resize_short
-
-    def __call__(self, img):
-        img_h, img_w = img.shape[:2]
-        percent = float(self.resize_short) / min(img_w, img_h)
-        w = int(round(img_w * percent))
-        h = int(round(img_h * percent))
-        return cv2.resize(img, (w, h))
-
-
-class CropImage(object):
-    def __init__(self, size):
-        if type(size) is int:
-            self.size = (size, size)
-        else:
-            self.size = size
-
-    def __call__(self, img):
-        w, h = self.size
-        img_h, img_w = img.shape[:2]
-        w_start = (img_w - w) // 2
-        h_start = (img_h - h) // 2
-
-        w_end = w_start + w
-        h_end = h_start + h
-        return img[h_start:h_end, w_start:w_end, :]
-
-
-class NormalizeImage(object):
-    def __init__(self, scale=None, mean=None, std=None):
-        self.scale = np.float32(scale if scale is not None else 1.0 / 255.0)
-        mean = mean if mean is not None else [0.485, 0.456, 0.406]
-        std = std if std is not None else [0.229, 0.224, 0.225]
-
-        shape = (1, 1, 3)
-        self.mean = np.array(mean).reshape(shape).astype('float32')
-        self.std = np.array(std).reshape(shape).astype('float32')
-
-    def __call__(self, img):
-        return (img.astype('float32') * self.scale - self.mean) / self.std
-
-
-class ToTensor(object):
-    def __init__(self):
-        pass
-
-    def __call__(self, img):
-        img = img.transpose((2, 0, 1))
-        return img

docs/en/tutorials/quick_start_recognition_en.md

@@ -90,13 +90,13 @@ wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/infere
 cd ..
 
 # Download the demo data and unzip it
-wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/recognition_demo_data_en_v1.0.tar && tar -xf recognition_demo_data_en_v1.0.tar
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/recognition_demo_data_en_v1.1.tar && tar -xf recognition_demo_data_en_v1.1.tar
 ```
 
-Once unpacked, the `recognition_demo_data_v1.0` folder should have the following file structure.
+Once unpacked, the `recognition_demo_data_v1.1` folder should have the following file structure.
 
 ```
-├── recognition_demo_data_v1.0
+├── recognition_demo_data_v1.1
 │   ├── gallery_cartoon
 │   ├── gallery_logo
 │   ├── gallery_product
@@ -126,13 +126,21 @@ The `models` folder should have the following file structure.
 <a name="Product_recognition_and_retrival"></a>
 ### 2.2 Product Recognition and Retrieval
 
-Take the product recognition demo as an example to show the recognition and retrieval process (if you wish to try other scenarios of recognition and retrieval, replace the corresponding configuration file after downloading and unzipping the corresponding demo data and model to complete the prediction)。
+Take the product recognition demo as an example to show the recognition and retrieval process (if you wish to try other scenarios of recognition and retrieval, replace the corresponding configuration file after downloading and unzipping the corresponding demo data and model to complete the prediction).
 
+**Note:**  `faiss` is used as search library. The installation method is as follows：
+
+```
+pip install faiss-cpu==1.7.1post2
+```
+
+If error happens when using `import faiss`, please uninstall `faiss` and reinstall it, especially on `Windows`.
 
 <a name="recognition_of_single_image"></a>
+
 #### 2.2.1 Single Image Recognition
 
-Run the following command to identify and retrieve the image `./recognition_demo_data_v1.0/test_product/daoxiangcunjinzhubing_6.jpg` for recognition and retrieval
+Run the following command to identify and retrieve the image `./recognition_demo_data_v1.1/test_product/daoxiangcunjinzhubing_6.jpg` for recognition and retrieval
 
 ```shell
 # use the following command to predict using GPU.
@@ -141,8 +149,6 @@ python3.7 python/predict_system.py -c configs/inference_product.yaml
 python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.use_gpu=False
 ```
 
-**Note:** Program lib used to build index is compliled on our machine, if error occured because of the environment, you can refer to [vector search tutorial](../../../deploy/vector_search/README.md) to rebuild the lib.
-
 
 The image to be retrieved is shown below.
 
@@ -175,7 +181,7 @@ If you want to predict the images in the folder, you can directly modify the `Gl
 
 ```shell
 # using the following command to predict using GPU, you can append `-o Global.use_gpu=False` to predict using CPU.
-python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.0/test_product/"
+python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.1/test_product/"
 ```
 
 
@@ -195,16 +201,16 @@ The results on the screen are shown as following.
 All the visualization results are also saved in folder `output`.
 
 
-Furthermore, the recognition inference model path can be changed by modifying the `Global.rec_inference_model_dir` field, and the path of the index to the index databass can be changed by modifying the `IndexProcess.index_path` field.
+Furthermore, the recognition inference model path can be changed by modifying the `Global.rec_inference_model_dir` field, and the path of the index to the index databass can be changed by modifying the `IndexProcess.index_dir` field.
 
 
 <a name="unkonw_category_image_recognition_experience"></a>
 ## 3. Recognize Images of Unknown Category
 
-To recognize the image `./recognition_demo_data_v1.0/test_product/anmuxi.jpg`, run the command as follows:
+To recognize the image `./recognition_demo_data_v1.1/test_product/anmuxi.jpg`, run the command as follows:
 
 ```shell
-python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.0/test_product/anmuxi.jpg"
+python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.1/test_product/anmuxi.jpg"
 ```
 
 The image to be retrieved is shown below.
@@ -225,14 +231,14 @@ When the index database cannot cover the scenes we actually recognise, i.e. when
 First, you need to copy the images which are similar with the image to retrieval to the original images for the index database. The command is as follows.
 
 ```shell
-cp -r  ../docs/images/recognition/product_demo/gallery/anmuxi ./recognition_demo_data_v1.0/gallery_product/gallery/
+cp -r  ../docs/images/recognition/product_demo/gallery/anmuxi ./recognition_demo_data_/gallery_product/gallery/
 ```
 
 Then you need to create a new label file which records the image path and label information. Use the following command to create a new file based on the original one.
 
 ```shell
 # copy the file
-cp recognition_demo_data_v1.0/gallery_product/data_file.txt recognition_demo_data_v1.0/gallery_product/data_file_update.txt
+cp recognition_demo_data_v1.1/gallery_product/data_file.txt recognition_demo_data_v1.1/gallery_product/data_file_update.txt
 ```
 
 Then add some new lines into the new label file, which is shown as follows.
@@ -255,20 +261,20 @@ Each line can be splited into two fields. The first field denotes the relative i
 Use the following command to build the index to accelerate the retrieval process after recognition.
 
 ```shell
-python3.7 python/build_gallery.py -c configs/build_product.yaml -o IndexProcess.data_file="./recognition_demo_data_v1.0/gallery_product/data_file_update.txt" -o IndexProcess.index_path="./recognition_demo_data_v1.0/gallery_product/index_update"
+python3.7 python/build_gallery.py -c configs/build_product.yaml -o IndexProcess.data_file="./recognition_demo_data_v1.1/gallery_product/data_file_update.txt" -o IndexProcess.index_dir="./recognition_demo_data_v1.1/gallery_product/index_update"
 ```
 
-Finally, the new index information is stored in the folder`./recognition_demo_data_v1.0/gallery_product/index_update`. Use the new index database for the above index.
+Finally, the new index information is stored in the folder`./recognition_demo_data_v1.1/gallery_product/index_update`. Use the new index database for the above index.
 
 
 <a name="Image_differentiation_based_on_the_new_index_library"></a>
 ### 3.2 Recognize the Unknown Category Images
 
-To recognize the image `./recognition_demo_data_v1.0/test_product/anmuxi.jpg`, run the command as follows.
+To recognize the image `./recognition_demo_data_v1.1/test_product/anmuxi.jpg`, run the command as follows.
 
 ```shell
 # using the following command to predict using GPU, you can append `-o Global.use_gpu=False` to predict using CPU.
-python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.0/test_product/anmuxi.jpg" -o IndexProcess.index_path="./recognition_demo_data_v1.0/gallery_product/index_update"
+python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.1/test_product/anmuxi.jpg" -o IndexProcess.index_dir="./recognition_demo_data_v1.1/gallery_product/index_update"
 ```
 
 The output is as follows:

docs/zh_CN/tutorials/quick_start_recognition.md

@@ -46,8 +46,8 @@
 
 本章节demo数据下载地址如下: [数据下载链接](https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/recognition_demo_data_v1.1.tar)。
 
-
 **注意**
+
 1. windows 环境下如果没有安装wget,可以按照下面的步骤安装wget与tar命令，也可以在，下载模型时将链接复制到浏览器中下载，并解压放置在相应目录下；linux或者macOS用户可以右键点击，然后复制下载链接，即可通过`wget`命令下载。
 2. 如果macOS环境下没有安装`wget`命令，可以运行下面的命令进行安装。
 
@@ -74,8 +74,8 @@ cd ..
 wget {数据下载链接地址} && tar -xf {压缩包的名称}
 ```
 
-
 <a name="下载、解压inference_模型与demo数据"></a>
+
 ### 2.1 下载、解压inference 模型与demo数据
 
 以商品识别为例，下载demo数据集以及通用检测、识别模型，命令如下。
@@ -90,13 +90,13 @@ wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/infere
 
 cd ../
 # 下载demo数据并解压
-wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/recognition_demo_data_v1.0.tar && tar -xf recognition_demo_data_v1.0.tar
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/recognition_demo_data_v1.1.tar && tar -xf recognition_demo_data_v1.1.tar
 ```
 
-解压完毕后，`recognition_demo_data_v1.0`文件夹下应有如下文件结构：
+解压完毕后，`recognition_demo_data_v1.1`文件夹下应有如下文件结构：
 
 ```
-├── recognition_demo_data_v1.0
+├── recognition_demo_data_v1.1
 │   ├── gallery_cartoon
 │   ├── gallery_logo
 │   ├── gallery_product
@@ -129,11 +129,19 @@ wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/recognit
 
 以商品识别demo为例，展示识别与检索过程（如果希望尝试其他方向的识别与检索效果，在下载解压好对应的demo数据与模型之后，替换对应的配置文件即可完成预测）。
 
+注意，此部分使用了`faiss`作为检索库，安装方法如下：
+
+```python
+pip install faiss-cpu==1.7.1post2
+```
+
+若使用时，不能正常引用，则`uninstall` 之后，重新`install`，尤其是windows下。
 
 <a name="识别单张图像"></a>
+
 #### 2.2.1 识别单张图像
 
-运行下面的命令，对图像`./recognition_demo_data_v1.0/test_product/daoxiangcunjinzhubing_6.jpg`进行识别与检索
+运行下面的命令，对图像`./recognition_demo_data_v1.1/test_product/daoxiangcunjinzhubing_6.jpg`进行识别与检索
 
 ```shell
 # 使用下面的命令使用GPU进行预测
@@ -142,8 +150,6 @@ python3.7 python/predict_system.py -c configs/inference_product.yaml
 python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.use_gpu=False
 ```
 
-注意：这里使用了默认编译生成的库文件进行特征索引，如果与您的环境不兼容，导致程序报错，可以参考[向量检索教程](../../../deploy/vector_search/README.md)重新编译库文件。
-
 待检索图像如下所示。
 
 <div align="center">
@@ -173,7 +179,7 @@ python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.u
 
 ```shell
 # 使用下面的命令使用GPU进行预测，如果希望使用CPU预测，可以在命令后面添加-o Global.use_gpu=False
-python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.0/test_product/"
+python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.1/test_product/"
 ```
 
 终端中会输出该文件夹内所有图像的识别结果，如下所示。
@@ -193,17 +199,17 @@ python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.i
 所有图像的识别结果可视化图像也保存在`output`文件夹内。
 
 
-更多地，可以通过修改`Global.rec_inference_model_dir`字段来更改识别inference模型的路径，通过修改`IndexProcess.index_path`字段来更改索引库索引的路径。
-
+更多地，可以通过修改`Global.rec_inference_model_dir`字段来更改识别inference模型的路径，通过修改`IndexProcess.index_dir`字段来更改索引库索引的路径。
 
 <a name="未知类别的图像识别体验"></a>
+
 ## 3. 未知类别的图像识别体验
 
-对图像`./recognition_demo_data_v1.0/test_product/anmuxi.jpg`进行识别，命令如下
+对图像`./recognition_demo_data_v1.1/test_product/anmuxi.jpg`进行识别，命令如下
 
 ```shell
 # 使用下面的命令使用GPU进行预测，如果希望使用CPU预测，可以在命令后面添加-o Global.use_gpu=False
-python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.0/test_product/anmuxi.jpg"
+python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.1/test_product/anmuxi.jpg"
 ```
 
 待检索图像如下所示。
@@ -222,20 +228,20 @@ python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.i
 <a name="准备新的数据与标签"></a>
 ### 3.1 准备新的数据与标签
 
-首先需要将与待检索图像相似的图像列表拷贝到索引库原始图像的文件夹(`./recognition_demo_data_v1.0/gallery_product/gallery`)中，运行下面的命令拷贝相似图像。
+首先需要将与待检索图像相似的图像列表拷贝到索引库原始图像的文件夹(`./recognition_demo_data_v1.1/gallery_product/gallery`)中，运行下面的命令拷贝相似图像。
 
 ```shell
-cp -r  ../docs/images/recognition/product_demo/gallery/anmuxi ./recognition_demo_data_v1.0/gallery_product/gallery/
+cp -r  ../docs/images/recognition/product_demo/gallery/anmuxi ./recognition_demo_data_v1.1/gallery_product/gallery/
 ```
 
-然后需要编辑记录了图像路径和标签信息的文本文件(`./recognition_demo_data_v1.0/gallery_product/data_file_update.txt`)，这里基于原始标签文件，新建一个文件。命令如下。
+然后需要编辑记录了图像路径和标签信息的文本文件(`./recognition_demo_data_v1.1/gallery_product/data_file_update.txt`)，这里基于原始标签文件，新建一个文件。命令如下。
 
 ```shell
 # 复制文件
-cp recognition_demo_data_v1.0/gallery_product/data_file.txt recognition_demo_data_v1.0/gallery_product/data_file_update.txt
+cp recognition_demo_data_v1.1/gallery_product/data_file.txt recognition_demo_data_v1.1/gallery_product/data_file_update.txt
 ```
 
-然后在文件`recognition_demo_data_v1.0/gallery_product/data_file_update.txt`中添加以下的信息，
+然后在文件`recognition_demo_data_v1.1/gallery_product/data_file_update.txt`中添加以下的信息，
 
 ```
 gallery/anmuxi/001.jpg	安慕希酸奶
@@ -255,20 +261,20 @@ gallery/anmuxi/006.jpg	安慕希酸奶
 使用下面的命令构建index索引，加速识别后的检索过程。
 
 ```shell
-python3.7 python/build_gallery.py -c configs/build_product.yaml -o IndexProcess.data_file="./recognition_demo_data_v1.0/gallery_product/data_file_update.txt" -o IndexProcess.index_path="./recognition_demo_data_v1.0/gallery_product/index_update"
+python3.7 python/build_gallery.py -c configs/build_product.yaml -o IndexProcess.data_file="./recognition_demo_data_v1.1/gallery_product/data_file_update.txt" -o IndexProcess.index_dir="./recognition_demo_data_v1.1/gallery_product/index_update"
 ```
 
-最终新的索引信息保存在文件夹`./recognition_demo_data_v1.0/gallery_product/index_update`中。
-
+最终新的索引信息保存在文件夹`./recognition_demo_data_v1.1/gallery_product/index_update`中。
 
 <a name="基于新的索引库的图像识别"></a>
+
 ### 3.3 基于新的索引库的图像识别
 
 使用新的索引库，对上述图像进行识别，运行命令如下。
 
 ```shell
 # 使用下面的命令使用GPU进行预测，如果希望使用CPU预测，可以在命令后面添加-o Global.use_gpu=False
-python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.0/test_product/anmuxi.jpg" -o IndexProcess.index_path="./recognition_demo_data_v1.0/gallery_product/index_update"
+python3.7 python/predict_system.py -c configs/inference_product.yaml -o Global.infer_imgs="./recognition_demo_data_v1.1/test_product/anmuxi.jpg" -o IndexProcess.index_dir="./recognition_demo_data_v1.1/gallery_product/index_update"
 ```
 
 输出结果如下。

docs/zh_CN_tmp/algorithm_introduction/metric_learning.md

+# Metric Learning
+
+## 简介
+   在机器学习中，我们经常会遇到度量数据间距离的问题。一般来说，对于可度量的数据，我们可以直接通过欧式距离(Euclidean Distance)，向量内积（Inner Product）或者是余弦相似度（Cosine Similarity）来进行计算。但对于非结构化数据来说，我们却很难进行这样的操作，如计算一段视频和一首音乐的匹配程度。由于数据格式的不同，我们难以直接进行上述的向量运算，但先验知识告诉我们ED(laugh_video, laugh_music) < ED(laugh_video, blue_music), 如何去有效得表征这种”距离”关系呢? 这就是Metric Learning所要研究的课题。
+   
+   Metric learning全称是 Distance Metric Learning，它是通过机器学习的形式，根据训练数据，自动构造出一种基于特定任务的度量函数。Metric Learning的目标是学习一个变换函数（线性非线性均可）L，将数据点从原始的向量空间映射到一个新的向量空间，在新的向量空间里相似点的距离更近，非相似点的距离更远，使得度量更符合任务的要求，如下图所示。 Deep Metric Learning，就是用深度神经网络来拟合这个变换函数。
+![example](../../images/ml_illustration.jpg)
+
+
+## 应用
+   Metric Learning技术在生活实际中应用广泛，如我们耳熟能详的人脸识别（Face Recognition）、行人重识别（Person ReID）、图像检索（Image Retrieval）、细粒度分类（Fine-gained classification）等.  随着深度学习在工业实践中越来越广泛的应用，目前大家研究的方向基本都偏向于Deep Metric Learning（DML). 
+   
+   一般来说, DML包含三个部分: 特征提取网络来map embedding, 一个采样策略来将一个mini-batch里的样本组合成很多个sub-set, 最后loss function在每个sub-set上计算loss. 如下图所示：
+   ![image](../../images/ml_pipeline.jpg)
+
+
+## 算法
+   Metric Learning主要有如下两种学习范式：
+### 1.	Classification based:  
+   这是一类基于分类标签的Metric Learning方法。这类方法通过将每个样本分类到正确的类别中，来学习有效的特征表示，学习过程中需要每个样本的显式标签参与Loss计算。常见的算法有[L2-Softmax](https://arxiv.org/abs/1703.09507), [Large-margin Softmax](https://arxiv.org/abs/1612.02295), [Angular Softmax](https://arxiv.org/pdf/1704.08063.pdf), [NormFace](https://arxiv.org/abs/1704.06369), [AM-Softmax](https://arxiv.org/abs/1801.05599), [CosFace](https://arxiv.org/abs/1801.09414), [ArcFace](https://arxiv.org/abs/1801.07698)等。 
+   这类方法也被称作是proxy-based, 因为其本质上优化的是样本和一堆proxies之间的相似度。
+### 2.	Pairwise based: 
+   这是一类基于样本对的学习范式。他以样本对作为输入，通过直接学习样本对之间的相似度来得到有效的特征表示，常见的算法包括：[Contrastive loss](http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf), [Triplet loss](https://arxiv.org/abs/1503.03832), [Lifted-Structure loss](https://arxiv.org/abs/1511.06452), [N-pair loss](https://papers.nips.cc/paper/2016/file/6b180037abbebea991d8b1232f8a8ca9-Paper.pdf), [Multi-Similarity loss](https://arxiv.org/pdf/1904.06627.pdf)等
+
+2020年发表的[CircleLoss](https://arxiv.org/abs/2002.10857)，从一个全新的视角统一了两种学习范式，让研究人员和从业者对Metric Learning问题有了更进一步的思考。
+

requirements.txt

@@ -8,4 +8,4 @@ visualdl >= 2.0.0b
 scipy
 scikit-learn==0.23.2
 gast==0.3.3
-faiss-cpu==1.7.1
+faiss-cpu==1.7.1.post2

tests/ResNet50_vd.txt → tests/config/ResNet50_vd.txt

@@ -49,3 +49,10 @@ inference:python/predict_cls.py -c configs/inference_cls.yaml
 -o Global.save_log_path:null
 -o Global.benchmark:True
 null:null
+null:null
+===========================cpp_infer_params===========================
+use_gpu:0|1
+cpu_threads:1|6
+use_mkldnn:0|1
+use_tensorrt:0|1
+use_fp16:0|1

tests/config/cpp_config.txt

+# model load config
+gpu_id 0
+gpu_mem 2000
+
+# whole chain test will add following config
+# use_gpu 0
+# cpu_threads 10
+# use_mkldnn 1
+# use_tensorrt 0
+# use_fp16 0
+
+# cls config
+cls_model_path ../inference/inference.pdmodel
+cls_params_path ../inference/inference.pdiparams
+resize_short_size 256
+crop_size 224
+
+# for log env info
+benchmark 1

tests/prepare.sh

@@ -33,7 +33,7 @@ if [ ${MODE} = "lite_train_infer" ] || [ ${MODE} = "whole_infer" ];then
     mv train.txt train_list.txt
     mv val.txt val_list.txt
     cd ../../
-elif [ ${MODE} = "infer" ];then
+elif [ ${MODE} = "infer" ] || [ ${MODE} = "cpp_infer" ];then
     # download data
     cd dataset
     rm -rf ILSVRC2012
@@ -58,3 +58,63 @@ elif [ ${MODE} = "whole_train_infer" ];then
     mv val.txt val_list.txt
     cd ../../
 fi
+
+if [ ${MODE} = "cpp_infer" ];then
+    cd deploy/cpp
+    echo "################### build opencv ###################"
+    rm -rf 3.4.7.tar.gz opencv-3.4.7/
+    wget https://github.com/opencv/opencv/archive/3.4.7.tar.gz
+    tar -xf 3.4.7.tar.gz
+    install_path=$(pwd)/opencv-3.4.7/opencv3
+    cd opencv-3.4.7/
+
+    rm -rf build
+    mkdir build
+    cd build
+    cmake .. \
+	-DCMAKE_INSTALL_PREFIX=${install_path} \
+	-DCMAKE_BUILD_TYPE=Release \
+	-DBUILD_SHARED_LIBS=OFF \
+	-DWITH_IPP=OFF \
+	-DBUILD_IPP_IW=OFF \
+	-DWITH_LAPACK=OFF \
+	-DWITH_EIGEN=OFF \
+	-DCMAKE_INSTALL_LIBDIR=lib64 \
+	-DWITH_ZLIB=ON \
+	-DBUILD_ZLIB=ON \
+	-DWITH_JPEG=ON \
+	-DBUILD_JPEG=ON \
+	-DWITH_PNG=ON \
+	-DBUILD_PNG=ON \
+	-DWITH_TIFF=ON \
+	-DBUILD_TIFF=ON
+     make -j
+     make install
+     cd ../../
+     echo "################### build opencv finished ###################"
+
+     echo "################### build PaddleClas demo ####################"
+     OPENCV_DIR=$(pwd)/opencv-3.4.7/opencv3/
+     LIB_DIR=$(pwd)/Paddle/build/paddle_inference_install_dir/
+     CUDA_LIB_DIR=$(dirname `find /usr -name libcudart.so`)
+     CUDNN_LIB_DIR=$(dirname `find /usr -name libcudnn.so`)
+
+     BUILD_DIR=build
+     rm -rf ${BUILD_DIR}
+     mkdir ${BUILD_DIR}
+     cd ${BUILD_DIR}
+     cmake .. \
+        -DPADDLE_LIB=${LIB_DIR} \
+        -DWITH_MKL=ON \
+        -DDEMO_NAME=clas_system \
+        -DWITH_GPU=OFF \
+        -DWITH_STATIC_LIB=OFF \
+        -DWITH_TENSORRT=OFF \
+        -DTENSORRT_DIR=${TENSORRT_DIR} \
+        -DOPENCV_DIR=${OPENCV_DIR} \
+        -DCUDNN_LIB=${CUDNN_LIB_DIR} \
+        -DCUDA_LIB=${CUDA_LIB_DIR} \
+
+     make -j
+     echo "################### build PaddleClas demo finished ###################"
+fi

tests/test.sh

 #!/bin/bash
 FILENAME=$1
-# MODE be one of ['lite_train_infer' 'whole_infer' 'whole_train_infer', 'infer'] 
+# MODE be one of ['lite_train_infer' 'whole_infer' 'whole_train_infer', 'infer', 'cpp_infer'] 
 MODE=$2
 
 dataline=$(cat ${FILENAME})
@@ -145,10 +145,80 @@ benchmark_value=$(func_parser_value "${lines[49]}")
 infer_key1=$(func_parser_key "${lines[50]}")
 infer_value1=$(func_parser_value "${lines[50]}")
 
+if [ ${MODE} = "cpp_infer" ]; then
+    cpp_use_gpu_key=$(func_parser_key "${lines[53]}")
+    cpp_use_gpu_list=$(func_parser_value "${lines[53]}")
+    cpp_cpu_threads_key=$(func_parser_key "${lines[54]}")
+    cpp_cpu_threads_list=$(func_parser_value "${lines[54]}")
+    cpp_use_mkldnn_key=$(func_parser_key "${lines[55]}")
+    cpp_use_mkldnn_list=$(func_parser_value "${lines[55]}")
+    cpp_use_tensorrt_key=$(func_parser_key "${lines[56]}")
+    cpp_use_tensorrt_list=$(func_parser_value "${lines[56]}")
+    cpp_use_fp16_key=$(func_parser_key "${lines[57]}")
+    cpp_use_fp16_list=$(func_parser_value "${lines[57]}")
+fi
+
 LOG_PATH="./tests/output"
 mkdir -p ${LOG_PATH}
 status_log="${LOG_PATH}/results.log"
 
+function func_cpp_inference(){
+    IFS='|'
+    _script=$1
+    _log_path=$2
+    _img_dir=$3
+    # inference 
+    for use_gpu in ${cpp_use_gpu_list[*]}; do
+        if [ ${use_gpu} = "0" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${cpp_use_mkldnn_list[*]}; do
+                if [ ${use_mkldnn} = "0" ] && [ ${_flag_quant} = "True" ]; then
+                    continue
+                fi
+                for threads in ${cpp_cpu_threads_list[*]}; do
+                    _save_log_path="${_log_path}/cpp_infer_cpu_usemkldnn_${use_mkldnn}_threads_${threads}.log"
+                    set_infer_data=$(func_set_params "${cpp_image_dir_key}" "${_img_dir}")
+		    cp ../tests/config/cpp_config.txt cpp_config.txt
+		    echo "${cpp_use_gpu_key} ${use_gpu}" >> cpp_config.txt
+		    echo "${cpp_cpu_threads_key} ${threads}" >> cpp_config.txt
+		    echo "${cpp_use_mkldnn_key} ${use_mkldnn}" >> cpp_config.txt
+		    echo "${cpp_use_tensorrt_key} 0" >> cpp_config.txt
+		    echo "${cpp_use_fp16_key} 0" >> cpp_config.txt
+                    command="${_script} cpp_config.txt ${_img_dir} > ${_save_log_path} 2>&1 "
+                    eval $command
+                    last_status=${PIPESTATUS[0]}
+                    eval "cat ${_save_log_path}"
+                    status_check $last_status "${command}" "${status_log}"
+                done
+            done
+        elif [ ${use_gpu} = "1" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${cpp_use_tensorrt_list[*]}; do
+                for precision in ${cpp_use_fp16_list[*]}; do
+                    if [[ ${precision} =~ "fp16" || ${precision} =~ "int8" ]] && [ ${use_trt} = "False" ]; then
+                        continue
+                    fi
+                    if [[ ${use_trt} = "False" || ${precision} =~ "int8" ]] && [ ${_flag_quant} = "True" ]; then
+                        continue
+                    fi
+                    _save_log_path="${_log_path}/cpp_infer_gpu_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+		    cp ../tests/config/cpp_config.txt cpp_config.txt
+		    echo "${cpp_use_gpu_key} ${use_gpu}" >> cpp_config.txt
+		    echo "${cpp_cpu_threads_key} ${threads}" >> cpp_config.txt
+		    echo "${cpp_use_mkldnn_key} ${use_mkldnn}" >> cpp_config.txt
+		    echo "${cpp_use_tensorrt_key} ${use_trt}" >> cpp_config.txt
+		    echo "${cpp_use_fp16_key} ${precision}" >> cpp_config.txt
+                    command="${_script} cpp_config.txt ${_img_dir} > ${_save_log_path} 2>&1 "
+                    eval $command
+                    last_status=${PIPESTATUS[0]}
+                    eval "cat ${_save_log_path}"
+                    status_check $last_status "${command}" "${status_log}"
+                        
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
 
 function func_inference(){
     IFS='|'
@@ -247,6 +317,10 @@ if [ ${MODE} = "infer" ]; then
         Count=$(($Count + 1))
     done
     cd ..
+elif [ ${MODE} = "cpp_infer" ]; then
+    cd deploy
+    func_cpp_inference "./cpp/build/clas_system" "../${LOG_PATH}" "${infer_img_dir}"
+    cd ..
 
 else
     IFS="|"