Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/Tsinghua-MARS-Lab/futr3d
Code for paper: FUTR3D: a unified sensor fusion framework for 3d detection
https://github.com/Tsinghua-MARS-Lab/futr3d
Last synced: 3 months ago
JSON representation
Code for paper: FUTR3D: a unified sensor fusion framework for 3d detection
- Host: GitHub
- URL: https://github.com/Tsinghua-MARS-Lab/futr3d
- Owner: Tsinghua-MARS-Lab
- License: apache-2.0
- Created: 2022-06-04T08:21:17.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2023-07-06T05:50:45.000Z (over 1 year ago)
- Last Synced: 2024-08-01T03:43:30.868Z (5 months ago)
- Language: Python
- Homepage:
- Size: 92.9 MB
- Stars: 266
- Watchers: 16
- Forks: 39
- Open Issues: 26
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Citation: CITATION.cff
Awesome Lists containing this project
- Awesome-BEV-Perception - project
README
# FUTR3D: A Unified Sensor Fusion Framework for 3D Detection
This repo implements the paper FUTR3D: A Unified Sensor Fusion Framework for 3D Detection. [Paper](https://arxiv.org/abs/2203.10642) - [project page](https://tsinghua-mars-lab.github.io/futr3d/)
We built our implementation upon MMdetection3D 1.0.0rc6. The major part of the code is in the directory `plugin/futr3d`.
## Environment
### Prerequisite
- mmcv-full>=1.5.2, <=1.7.0
- mmdet>=2.24.0, <=3.0.0
- mmseg>=0.20.0, <=1.0.0
- nuscenes-devkit
### Installation
There is no neccesary to install mmdet3d separately, please install based on this repo:
```
cd futr3d
pip3 install -v -e .
```
### Data
Please follow the mmdet3d to process the data. [mmdet3d_nuscenes_guidance](https://github.com/open-mmlab/mmdetection3d/blob/main/docs/en/advanced_guides/datasets/nuscenes.md)
Notably, we have modified the nuscenes_converter.py to add the radar infomation, so the infos.pkl generated by our code is different from the original code. The other infomation except the radar infos is the same with the original infos.pkl.
## Train
For example, to train FUTR3D with LiDAR only on 8 GPUs, please use
```
bash tools/dist_train.sh plugin/futr3d/configs/lidar_only/lidar_0075_900q.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](https://github.com/WangYueFt/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 --lidar --out
```
to fuse cam-only and lidar-only models.
## Evaluate
For example, to evalaute FUTR3D with LiDAR-cam on 8 GPUs, please use
```
bash tools/dist_train.sh plugin/futr3d/configs/lidar_cam/lidar_0075_cam_res101.py ../lidar_cam.pth 8 --eval bbox
```
## Results
### LiDAR & Cam
| models | mAP | NDS | Link |
| ----------- | ----------- | ----| ---- |
| [Res101 + VoxelNet](./plugin/futr3d/configs/lidar_cam/lidar_0075v_cam_res101.py) | 67.4 | 70.9 | [model](https://drive.google.com/file/d/1hdsrQhWOD6CjgoTgyi1i3KV94IRt2OhF/view?usp=share_link)|
| [VoVNet + VoxelNet](./plugin/futr3d/configs/lidar_cam/lidar_0075v_cam_vov.py) | 70.3 | 73.1 | [model](https://drive.google.com/file/d/1DgrzSoZSlTT_RDNGplHUMXatboKlkCqq/view?usp=share_link) |
### Cam & Radar
| models | mAP | NDS | Link |
| ----------- | ----------- | ----| ----- |
| [Res101 + Radar](./plugin/futr3d/configs/cam_radar/cam_res101_radar.py) | 39.9 | 50.8 | [model](https://drive.google.com/file/d/1LX3kflWap_qWjTNy3Zy9gL9_IXANkUo1/view?usp=share_link) |
### LiDAR only
| models | mAP | NDS | Link |
| ----------- | ----------- | ----| ----|
| [32 beam VoxelNet](./plugin/futr3d/configs/lidar_only/lidar_0075v_900q.py) | 63.3 | 68.9 | [model](https://drive.google.com/file/d/1660B8m1CsDf_DwxdD_sXsdSTL7FbBnZn/view?usp=share_link)|
| [4 beam VoxelNet](./plugin/futr3d/configs/lidar_only/lidar_0075v_900q_4b.py) | 44.3 | 56.4 |
| [1 beam VoxelNet](./plugin/futr3d/configs/lidar_only/lidar_0075v_900q_1b.py) | 16.9 | 39.2 |
### Cam only
The camera-only version of FUTR3D is the same as DETR3D. Please check [DETR3D](https://github.com/WangYueFt/detr3d) for detail implementation.
## Acknowledgment
For the implementation, we rely heavily on [MMCV](https://github.com/open-mmlab/mmcv), [MMDetection](https://github.com/open-mmlab/mmdetection), [MMDetection3D](https://github.com/open-mmlab/mmdetection3d), and [DETR3D](https://github.com/WangYueFt/detr3d)
## Related projects
1. [DETR3D: 3D Object Detection from Multi-view Images via 3D-to-2D Queries](https://tsinghua-mars-lab.github.io/detr3d/)
2. [MUTR3D: A Multi-camera Tracking Framework via 3D-to-2D Queries](https://tsinghua-mars-lab.github.io/mutr3d/)
3. For more projects on Autonomous Driving, check out our Visual-Centric Autonomous Driving (VCAD) project page [webpage](https://tsinghua-mars-lab.github.io/vcad/)
## Reference
```
@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]`