Pipeline Generation in Custom Scenarios

Key Points

1. Define the function generated by the inference service.

   get_pipeline_name: returns the main function name of the process orchestration. Execute the main function script infer_service_generation.py of the inference service when a custom scenario is scanned, and bind the main function name of the custom process orchestration to the name of the custom scenario.

   Main function for process orchestration: orchestrates service processes in custom scenarios. Invoke the main function for process orchestration based on the preceding bound mapping.

2. Orchestrate the process using mxManufacture.

   Process orchestration defines the model serial connections in the scenario-based service process. The process orchestrated by mxManufacture helps quickly develop high-performance applications.

   See the following example.

   from mindx.sdk.stream import PluginNode, FunctionalStream
    
   if __name__ == '__main__':
       # Define the attributes of the mxpi_tensorinfer plugin.
       props0 = {
           "dataSource": "mxpi_imageresize0",
           "modelPath": "./ssd_mobilenetv1_fpn_best.om" # Use the actual model path.
   }
   # Define the attributes of the mxpi_objectpostprocessor plugin.
       props1 = {
           "dataSource": "mxpi_tensorinfer0",
           "postProcessConfigPath": "./post_process_0.cfg", # Use the actual configuration file.
           "labelPath": "./label_0.names", # Use the actual configuration file.
           "postProcessLibPath": "libSsdMobilenetFpn_MindsporePost.so"
   }
   # Define the attributes of the mxpi_objectpostprocessor plugin.
       props2 = {
           "outputDataKeys":"mxpi_objectpostprocessor0"
       } 
       # Connects models in series 
       appsrc0 = PluginNode("appsrc")
       mxpi_imagedecoder0 = PluginNode("mxpi_imagedecoder")(appsrc0)
       mxpi_imageresize0 = PluginNode("mxpi_imageresize")(mxpi_imagedecoder0)
       mxpi_tensorinfer0 = PluginNode("mxpi_tensorinfer", props0)(mxpi_imageresize0)
       mxpi_objectpostprocessor0 = PluginNode("mxpi_objectpostprocessor", props1)(mxpi_tensorinfer0)
       mxpi_dataserialize0 = PluginNode("mxpi_dataserialize", props2)(mxpi_objectpostprocessor0)
       appsink0 = PluginNode("appsink")(mxpi_dataserialize0)
       # Generate a pipeline.
       f=FunctionalStream('stream', [appsrc0], [appsink0])
       f.set_device_id('0')
       f.build()
       print('json pipeline:' + f.to_json())

3. Archive the inference model and configuration file. The path for archiving the inference model and configuration file should comply with the result_path field in the project configuration JSON file delivered by the platform.

Debugging Methods

After the custom inference service package is installed, run the following command:

python mxAOIService/scripts/infer_service_generation.py --model_params=path_of_the_project_configuration_JSON_file_delivered_by_the_platform

--model_params indicates the path of the project configuration JSON file delivered by the platform.

1. Check whether the mxAOIService/config/models/model_configs.json file is complete.
2. Check whether the mxAOIService/config/models/project_*/assm1-2/1/ model is ready.
3. Check whether the pipeline of mxAOIService/config/models/project_*/assm1-2/1/ is complete.
4. Check whether the post-processing configuration file of mxAOIService/config/models/project_*/assm1-2/1/ is complete.