This folder provides typical use cases in companion with the elegant mmdet3d package.
The example is based on the mmdet3d 1.0.0rc6, which is compatible with most popular BEV perception algorithms, like DETR3D, PETR, StreamPETR, BEVFormer, etc.
Please install mmdet3d following the official instructions.
Note that higher version (mmdet3d>1.0.0) is not tested, the given example may fail to run.
The file examples/mmdet3d/mmdet3d_plugin/datasets/roscenes_dataset.py shows how to implement a mmdet3d-compatible dataset using Custom3DDataset.
To use it, you need to declare with key in the training or evaluation configs. An example config is provided as examples/mmdet3d/configs/detr3d_roscene_res101.py, which should be compatible with DETR3D model.
An evaluation code is delivered together with roscenes_dataset.py. You can find it in RoSceneDataset.evaluate() function.
A detailed guide for performing evaluation is placed under roscenes/evaluation/README.md.