dazyoung / mrcnn_serving_ready Goto Github PK

Script to convert MatterPort Mask_RCNN Keras model to Tensorflow Frozen Graph and Tensorflow Serving Model.
Plus inferencing with GRPC or RESTAPI using Tensorflow Model Server.

1. Modify the path variables in 'user_config.py'
2. Run main.py

   python3 main.py

If you have a different config class you can replace the existing config in 'main.py'

# main.py
# Current config load
config = get_config()

# replace it with your config class
config = your_custom_config_class

Follow once you finish converting it to a saved_model using the above code

1. First run your saved_model.pb in Tensorflow Model Server, using:

   tensorflow_model_server --port=8500 --rest_api_port=8501 --model_name=mask --model_base_path=/path/to/saved_model/

2. Modify the variables and add your Config Class if needed in inferencing/saved_model_config.py. No need to change if the saved_model is the default COCO model.
3. Then run the inferencing/saved_model_inference.py with the image path:

   # Set Python Path
   export PYTHONPATH=$PYTHONPATH:$pwd
   
   # Run Inference with GRPC
   python3 inferencing/saved_model_inference.py -t grpc -p test_image/monalisa.jpg
   
   # Run Inference with RESTAPI
   python3 inferencing/saved_model_inference.py -t restapi -p test_image/monalisa.jpg

Thanks to @rahulgullan for RESTAPI client code.