How owen-liuyuxuan/visualdet3d works
Project type and purpose
• This is a research‐oriented, full‐pipeline Python toolkit for 3D object detection and depth estimation on the KITTI dataset.
File structure
• Top‐level:
• config/ – user‐editable .py configurations (data paths, hyperparameters)
• scripts/ & launchers/ – CLI entrypoints for preprocessing, training, evaluation
• make.sh – builds custom ops under visualDet3D/networks/lib/ops/
• demos/, docs/ – notebooks and documentation
• Core library: visualDet3D/ subdivided into:
• data/ – dataset loaders, samplers, augmentation pipelines
• networks/ – backbones, detectors, heads, loss modules, optimizers, training/evaluation pipelines
• evaluator/ – KITTI‐style metric calculators
• utils/ – common utilities (paths, timers, registry)
README
• The registry pattern allows plugging new datasets, models, transforms.
• Modular design supports mono and stereo 3D detection, depth prediction, multi‐GPU.
• Custom operations (deformable conv, IoU3D) built via make.sh.