Inference Handler Documentation

Handling Runs

handle_run.py

This code parses the JSON configuration file given by the user and executes commands per the available arguments in the configuration file. Specifically, the code block is concerned with arguments for:

  • Running pre model optimization, which is important when OpenVINO does not support the current model framework.

  • Running OpenVINO model optimization

  • Conducting inference on the data with the specified model/processors.

Functions in handle_run.py:

core.python.handle_run.get_params_dict(filename)

This function turns a filename string, opens and loads a JSON object as a dictionary, and returns this data dictionary.

Parameters

filename – The file name of a JSON file, most likely the configuration file.

Returns

Param Dictionary (data)

core.python.handle_run.json_args()

Parses the arguments passed in by the user. The only argument needed, in this case, is the file name to the configuration JSON file.

Param

None

Returns

Parsed Args (parser.parse_args())

Handling Inference

infer.py

This code sets up the inference task, and based on the arguments given, it will execute the task specific by the user in their configuration JSON file. Current inference tasks include:

  • Regular sync/async: This inference task is for more general purpose inference.

  • Fine Cube sync/async: This inference task is designed for salt identification tasks or tasks which involve taking in a mini cube of a larger input and outputting a unit volume to store in a larger cube. There is an option to interpolate the storage cube to the size of the input.

  • Coarse Cube sync: This inference task is designed for fault segmentation or tasks which involve taking in a mini cube of a larger input and outputting a mini cube of the same size to be stored in a larger storage cube.

  • Section sync/async: this inference task is designed for facies classification or tasks which involve conducting inference over sections of a larger sheet.

Functions in infer.py: