同步操作将从 PaddlePaddle/PaddleSeg 强制同步,此操作会覆盖自 Fork 仓库以来所做的任何修改,且无法恢复!!!
确定后同步将在后台操作,完成时将刷新页面,请耐心等待。
English|简体中文
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.
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
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
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.
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
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.
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。