DETR3D for mmdet3d-v1.0.0rc5 version

This repo is cloned from the original repo(https://github.com/WangYueFt/detr3d/).

This repo contains the implementations of DETR3D (https://arxiv.org/abs/2110.06922). The original DETR3D is released based on mmdet3d-0.17.0 version, however, the mmdet3d reconfigured the coordinate system in version 1.0, which caused problems with some metrics such as mAOE. Refer to issue31(WangYueFt/detr3d#31) we refactored this code and reproduced the results as in the paper.

Prerequisite

1. mmcv (https://github.com/open-mmlab/mmcv)

2. mmdet (https://github.com/open-mmlab/mmdetection)

3. mmseg (https://github.com/open-mmlab/mmsegmentation)

4. mmdet3d (https://github.com/open-mmlab/mmdetection3d)

Enviornment for mmdet3d and upper

  mmcv-full=1.6.0
  mmdet=2.24.0
  mmseg=0.20.0
  mmdet3d=1.0.0rc5
  pytorch=1.10.1

Data

1. Follow the mmdet3d to process the data.

Train

1. Downloads the pretrained backbone weights to pretrained.

2. For example, to train DETR3D on 8 GPUs, please use

tools/dist_train.sh projects/configs/detr3d/detr3d_res101_gridmask.py 8

Evaluation using pretrained models

1. Download the weights accordingly.

Backbone	mAP	NDS	Download
DETR3D, ResNet101 w/ DCN	34.7	42.2	model | log
above, + CBGS	34.9	43.4	model | log
DETR3D, VoVNet on trainval, evaluation on test set	41.2	47.9	model | log

2. To test, use
   tools/dist_test.sh /projects/configs/detr3d/detr3d_res101_gridmask.py /path/to/ckpt 8 --eval=bbox

Explanation of the changes in the coordinate system caused by the mmdet3d version update:

link: https://github.com/open-mmlab/mmdetection3d/blob/master/docs/en/compatibility.md#v100rc0

If you find this repo useful for your research, please consider citing the papers

@inproceedings{
   detr3d,
   title={DETR3D: 3D Object Detection from Multi-view Images via 3D-to-2D Queries},
   author={Wang, Yue and Guizilini, Vitor and Zhang, Tianyuan and Wang, Yilun and Zhao, Hang and and Solomon, Justin M.},
   booktitle={The Conference on Robot Learning ({CoRL})},
   year={2021}
}