• English
  • 3.7.0

    • Executing the Script

    In HAT, the functions that you can use and modify directly mainly include tools and configs. The tools are mainly the core functional modules, including training and validation visualization, while configs mainly contains the options and parameters that can be configured during the execution of the functional modules.

    This tutorial provides you with a general execution paradigm for config and an introduction to the core functions and external interfaces of tools.

    The Execution Paradigm of config

    In most cases, for the execution of tools, a config input is needed, except for some tools related to datasets or single image visualization. Therefore, the general execution paradigm can be summarized as follows:

    python3 tools/${TOOLS} --config configs/${CONFIGS}

    Here we mainly introduce the core functions and external interfaces of the tools.

    Introduction to the Functions and Parameters of the tools Utility

    The current tools have some python scripts, each for a different function.

    train.py is a training tool with the following major parameters:

    ParameterDescription
    --stage {float, calibration, qat}Different training and prediction stages.
    --config CONFIG, -c CONFIGPath to the config file.
    --device-ids DEVICE_IDS, -ids DEVICE_IDSList of running GPUs.
    --dist-url DIST_URLServer address for multi-computer operations, auto by default.
    --launcher {torch}Launch mode for multi-computer operations.
    --pipeline-testWhether to run the pipeline test.
    --optsModify config options using the command-line.
    --opts-overwriteWhether to allow modify config.
    --levelDefault logging level for other rank except rank0.

    predict.py is a predicting tool with the following major parameters:

    ParameterDescription
    --stage {float, calibration, qat}Different training and prediction stages.
    --config CONFIG, -c CONFIGPath to the config file.
    --device-ids DEVICE_IDS, -ids DEVICE_IDSList of running GPUs.
    --dist-url DIST_URLServer address for multi-computer operations, auto by default.
    --backendThe backend of communication methods of multiple nodes or GPUs.
    --launcher {torch}Launch mode for multi-computer operations.
    --ckptThe ckpt file for predict model.
    --pipeline-testWhether to run the pipeline test.

    model_checker.py is a checker tool for checking model executable on the BPU.

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.

    validation_hbir.py is an accuracy validation tool that provides fixed-point accuracy and fully aligned results with the on-board situations, with the following major parameters:

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.
    --stage {align_bpu}Different prediction stages.

    calops.py is the network ops calculation tool, with the following major parameters:

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.
    --input-shapeInput shape.
    --methodUse a calculated approach.

    compile_perf_hbir.py is the compilation and performance tool, with the following major parameters:

    Attention

    If you want to deploy the model provided by the reference algorithm package on nash-p, nash-b, or nash-h, you need to pass the corresponding march parameter.

    ParameterDescription
    --config CONFIG, -c CONFIGDirectory of the config file.
    --opt {0,1,2}Compilation-time optimization options.
    --jobs JOBSNumber of threads for compilation.
    --model_path MODEL_PATHThe path for qat hbir.
    --march ${MARCH}Optional parameter, bpu march.

    infer_hbir.py is used to perform single image prediction, with the following major parameters:

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.
    --model-inputsOptional parameter, specifies the path for the input data. If not provided, it will look for the input_path parameter in the config file instead.
    --save-pathThe path where the visualization results are saved.
    --use-datasetOptional parameter, if --model-inputs is not provided, and the config file does not contain input_path, or no data can be found under the input_path directory, then this parameter must be set to True to indicate that the data should be loaded from the dataset specified in the config file.

    create_data.py is used to pre-process Kitti3D lidar dataset, with the following major parameters:

    ParameterDescription
    --datasetName of dataset.
    --root-dirPath of dataset.

    export_onnx.py is used to export the model to onnx (only for visualization and does not support inference), with the following major parameters:

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.

    export_hbir.py is used to export hbir model, with the following major parameters:

    Attention

    If you want to deploy the model provided by the reference algorithm package on nash-p, nash-b, or nash-h, you need to pass the corresponding march parameter.

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.
    --save-pathPath of save hbir model.
    --march ${MARCH}Optional parameter, bpu march.

    gen_camera_param_nusc.py is used to get camera internal and external parameters from nuscenes, with the following major parameters:

    ParameterDescription
    --data-pathPath of dataset.
    --save-pathPath of save output.
    --save-by-cityWhether saved according to the city.
    --versionThe version of dataset.

    gen_reference_points_nusc.py is used to get model input reference points from nuscenes, with the following major parameters:

    ParameterDescription
    --data-pathPath of dataset.
    --save-pathPath of save output.
    --save-by-cityWhether saved according to the city.
    --versionThe version of dataset.
    --config CONFIG, -c CONFIGPath to the config file.

    homography_generator.pyis used to get ego2img matrix, with the following major parameters:

    ParameterDescription
    --sensor2ego-translationTranslation matrix from sensor to ego coordinate system.
    --sensor2ego-rotationRotation matrix from sensor to ego coordinate system.
    --camera-intrinsiccamera intrinsic.
    --save-pathPath of save output.

    reference_points_generator.py is used to calculate model input reference points from homography matrix, with the following major parameters:

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.
    --homographyPath to the homography file。
    --save-pathPath of save output.

    quant_analysis.py is used to analysis qat training, with the following major parameters:

    ParameterDescription
    --config CONFIG, -c CONFIGPath to the config file.

    The datasets directory is for dataset-related packaging and visualization tools.