Tools Overview

In HAT, the main components you can directly use and modify are tools and configs. The tools directory contains core functional modules such as training, validation, and visualization, while configs provides configuration options and parameters for each module.

The source code for tools and configs can be found in the horizon_j6_open_explorer package under samples/ai_toolchain/horizon_model_train_sample/scripts/.

The table below briefly introduces the main functions of each tool in the tools directory for your reference:

Tool Name	Function Description
train.py	Model training tool.
predict.py	Model prediction tool.
model_checker.py	Checks if the model can run properly on BPU.
calops.py	Network computation statistics tool.
compile_perf_hbir.py	Compilation and performance analysis tool.
infer_hbir.py	Single image inference tool.
create_data.py	Preprocessing tool for Kitti3D lidar dataset.
export_onnx.py	Exports model to ONNX (for visualization only).
export_hbir.py	Exports model to HBIR.
gen_camera_param_nusc.py	Script to extract camera parameters from nuscenes.
gen_reference_points_nusc.py	Script to extract model input reference points from nuscenes.
homography_generator.py	Script to compute ego-to-image transformation matrix.
reference_points_generator.py	Script to compute input reference points from homography matrix.
quant_analysis.py	Quantization training analysis tool.
dataset_converters	Tools for data format conversion.
datasets	Dataset packaging and visualization tools.

#Tools Overview

Tools Overview