English|简体中文

Paddle Serving deployment

Overview

The model trained by PaddleSeg can be deployed as a service using Paddle Serving.

This turtorial introduces the deployment method using Paddle Serving. For more details, please refer to the document.

Environmental preparation

Environment preparations are required on the server side and the client side. Please refer to document for more details.

On the server side:

• Install PaddlePaddle (version>=2.0)

• Install paddle-serving-app (version>=0.6.0)

• Install paddle-serving-server or paddle-serving-server-gpu (version>=0.6.0)

  pip3 install paddle-serving-app==0.6.0
  
  # CPU
  pip3 install paddle-serving-server==0.6.0
  
  # Choose paddle-serving-server-gpu according to your GPU environment
  pip3 install paddle-serving-server-gpu==0.6.0.post102 #GPU with CUDA10.2 + TensorRT7
  pip3 install paddle-serving-server-gpu==0.6.0.post101 # GPU with CUDA10.1 + TensorRT6
  pip3 install paddle-serving-server-gpu==0.6.0.post11 # GPU with CUDA10.1 + TensorRT7

On the client side:

• Install paddle-serving-app (version>=0.6.0)

• Install paddle-serving-client (version>=0.6.0)

  pip3 install paddle-serving-app==0.6.0
  pip3 install paddle-serving-client==0.6.0

Prepare model and data

Download the sample model for testing. If you want to use other models, please refer to model export tool.

$ wget https://paddleseg.bj.bcebos.com/dygraph/demo/bisenet_demo_model.tar.gz
tar zxvf bisenet_demo_model.tar.gz

Download a picture from cityscape to test. If your model is trained on other datasets, please prepare test images by yourself.

$ wget https://paddleseg.bj.bcebos.com/dygraph/demo/cityscapes_demo.png

Convert model

Before Paddle Serving is deployed, we need to convert the prediction model. For details, please refer to the document.

On the client side, execute the following script to convert the sample model.

python -m paddle_serving_client.convert \
    --dirname ./bisenetv2_demo_model \
    --model_filename model.pdmodel \
    --params_filename model.pdiparams

After excuting the script, the "serving_server" folder in the current directory saves the server model and configuration, and the "serving_client" folder saves the client model and configuration.

Server Deployment

You can use paddle_serving_server.serve to start the RPC service, please refer to the document.

If you finish to prepare environment on server side, export the server model and serving_server file, execute the following command to start the service. We use port 9292 on the server side. The server ip can be inquired by hostname -i.

python -m paddle_serving_server.serve \
    --model serving_server \
    --thread 10 \
    --port 9292 \
    --ir_optim

Client request service

cd PaddleSeg/deploy/serving

Set the path of the serving_client file, the server-side ip and port, and the path of the test picture, and execute the following commands.

python test_serving.py \
    --serving_client_path path/to/serving_client \
    --serving_ip_port ip:port \
    --image_path path/to/image\

After the execution is complete, the divided image is saved in "result.png" in the current directory.