This repo implements the paper FUTR3D: A Unified Sensor Fusion Framework for 3D Detection. Paper - project page
We built our implementation upon MMdetection3D 0.17.3. The major part of the code is in the directory plugin/futr3d
.
- mmcv-full>=1.3.8, <=1.4.0
- mmdet>=2.14.0, <=3.0.0
- mmseg>=0.14.1, <=1.0.0
- nuscenes-devkit
For cameras with Radar setting, you should generate a meta file or say .pkl
file including Radar infos.
python3 tools/data_converter/nuscenes_converter_radar.py
For others, please follow the mmdet3d to process the data. https://mmdetection3d.readthedocs.io/en/stable/datasets/nuscenes_det.html
For example, to train FUTR3D with LiDAR only on 8 GPUs, please use
bash tools/dist_train.sh plugin/futr3d/configs/lidar_only/01voxel_q6_step_38e.py 8
For LiDAR-Cam and Cam-Radar version, we need pre-trained model.
The Cam-Radar uses DETR3D model as pre-trained model, please check DETR3D.
The LiDAR-Cam uses fused LiDAR-only and Cam-only model as pre-trained model. You can use
python tools/fuse_model.py --img <cam checkpoint path> --lidar <lidar checkpoint path> --out <out model path>
to fuse cam-only and lidar-only models.
For example, to evalaute FUTR3D with LiDAR-cam on 8 GPUs, please use
bash tools/dist_train.sh plugin/futr3d/configs/lidar_cam/res101_01voxel_step_3e.py ../lidar_cam.pth 8 --eval bbox
models | mAP | NDS | Link |
---|---|---|---|
Res101 + 32 beam VoxelNet | 64.2 | 68.0 | model |
Res101 + 4 beam VoxelNet | 54.9 | 61.5 | |
Res101 + 1 beam VoxelNet | 41.3 | 50.0 |
models | mAP | NDS | Link |
---|---|---|---|
Res101 + Radar | 35.0 | 45.9 | model |
models | mAP | NDS | Link |
---|---|---|---|
32 beam VoxelNet | 59.3 | 65.5 | model |
4 beam VoxelNet | 42.1 | 54.8 | |
1 beam VoxelNet | 16.4 | 37.9 |
The camera-only version of FUTR3D is the same as DETR3D. Please check DETR3D for detail implementation.
For the implementation, we rely heavily on MMCV, MMDetection, MMDetection3D, and DETR3D
- DETR3D: 3D Object Detection from Multi-view Images via 3D-to-2D Queries
- MUTR3D: A Multi-camera Tracking Framework via 3D-to-2D Queries
- For more projects on Autonomous Driving, check out our Visual-Centric Autonomous Driving (VCAD) project page webpage
@article{chen2022futr3d,
title={FUTR3D: A Unified Sensor Fusion Framework for 3D Detection},
author={Chen, Xuanyao and Zhang, Tianyuan and Wang, Yue and Wang, Yilun and Zhao, Hang},
journal={arXiv preprint arXiv:2203.10642},
year={2022}
}
Contact: Xuanyao Chen at: [email protected]
or [email protected]