- **(2024/5)** BEVFusion is integrated into NVIDIA [DeepStream](https://developer.nvidia.com/blog/nvidia-deepstream-7-0-milestone-release-for-next-gen-vision-ai-development/) for sensor fusion.
- **(2023/4)** BEVFusion ranks first on [Argoverse](https://eval.ai/web/challenges/challenge-page/1710/overview) 3D object detection leaderboard among all solutions.
- **(2023/1)** BEVFusion is accepted to ICRA 2023!
- **(2022/8)** BEVFusion ranks first on [Waymo](https://waymo.com/open/challenges/2020/3d-detection/) 3D object detection leaderboard among all solutions.
- **(2022/6)** BEVFusion ranks first on [nuScenes](https://nuscenes.org/tracking?externalData=all&mapData=all&modalities=Any) 3D object detection leaderboard among all solutions.
- **(2022/6)** BEVFusion ranks first on [nuScenes](https://nuscenes.org/object-detection?externalData=all&mapData=all&modalities=Any) 3D object detection leaderboard among all solutions.
Multi-sensor fusion is essential for an accurate and reliable autonomous driving system. Recent approaches are based on point-level fusion: augmenting the LiDAR point cloud with camera features. However, the camera-to-LiDAR projection throws away the semantic density of camera features, hindering the effectiveness of such methods, especially for semantic-oriented tasks (such as 3D scene segmentation). In this paper, we break this deeply-rooted convention with BEVFusion, an efficient and generic multi-task multi-sensor fusion framework. It unifies multi-modal features in the shared bird's-eye view (BEV) representation space, which nicely preserves both geometric and semantic information. To achieve this, we diagnose and lift key efficiency bottlenecks in the view transformation with optimized BEV pooling, reducing latency by more than **40x**. BEVFusion is fundamentally task-agnostic and seamlessly supports different 3D perception tasks with almost no architectural changes. It establishes the new state of the art on the nuScenes benchmark, achieving **1.3%** higher mAP and NDS on 3D object detection and **13.6%** higher mIoU on BEV map segmentation, with **1.9x** lower computation cost.
Here, BEVFusion only uses a single model without any test time augmentation. BEVFusion-TTA uses single model with test-time augmentation and no model ensembling is applied.
*Note*: The camera-only object detection baseline is a variant of BEVDet-Tiny with a much heavier view transformer and other differences in hyperparameters. Thanks to our [efficient BEV pooling](mmdet3d/ops/bev_pool) operator, this model runs fast and has higher mAP than BEVDet-Tiny under the same input resolution. Please refer to [BEVDet repo](https://github.com/HuangJunjie2017/BEVDet) for the original BEVDet-Tiny implementation. The LiDAR-only baseline is TransFusion-L.
We also provide a [Dockerfile](docker/Dockerfile) to ease environment setup. To get started with docker, please make sure that `nvidia-docker` is installed on your machine. After that, please execute the following command to build the docker image:
```bash
cd docker && docker build . -t bevfusion
```
We can then run the docker with the following command:
```bash
nvidia-docker run -it -v `pwd`/../data:/dataset --shm-size 16g bevfusion /bin/bash
```
We recommend the users to run data preparation (instructions are available in the next section) outside the docker if possible. Note that the dataset directory should be an absolute path. Within the docker, please run the following command to clone our repo and install custom CUDA extensions:
```bash
cd home && git clone https://github.com/mit-han-lab/bevfusion && cd bevfusion
python setup.py develop
```
You can then create a symbolic link `data` to the `/dataset` directory in the docker.
Please follow the instructions from [here](https://github.com/open-mmlab/mmdetection3d/blob/master/docs/en/datasets/nuscenes_det.md) to download and preprocess the nuScenes dataset. Please remember to download both detection dataset and the map extension (for BEV map segmentation). After data preparation, you will be able to see the following directory structure (as is indicated in mmdetection3d):
[CUDA-BEVFusion](https://github.com/NVIDIA-AI-IOT/Lidar_AI_Solution/tree/master/CUDA-BEVFusion): Best practice for TensorRT, which provides INT8 acceleration solutions and achieves 25fps on ORIN.
Q: Can we directly use the info files prepared by mmdetection3d?
A: We recommend re-generating the info files using this codebase since we forked mmdetection3d before their [coordinate system refactoring](https://github.com/open-mmlab/mmdetection3d/blob/master/docs/en/changelog.md).
## Acknowledgements
BEVFusion is based on [mmdetection3d](https://github.com/open-mmlab/mmdetection3d). It is also greatly inspired by the following outstanding contributions to the open-source community: [LSS](https://github.com/nv-tlabs/lift-splat-shoot), [BEVDet](https://github.com/HuangJunjie2017/BEVDet), [TransFusion](https://github.com/XuyangBai/TransFusion), [CenterPoint](https://github.com/tianweiy/CenterPoint), [MVP](https://github.com/tianweiy/MVP), [FUTR3D](https://arxiv.org/abs/2203.10642), [CVT](https://github.com/bradyz/cross_view_transformers) and [DETR3D](https://github.com/WangYueFt/detr3d).
Please also check out related papers in the camera-only 3D perception community such as [BEVDet4D](https://arxiv.org/abs/2203.17054), [BEVerse](https://arxiv.org/abs/2205.09743), [BEVFormer](https://arxiv.org/abs/2203.17270), [M2BEV](https://arxiv.org/abs/2204.05088), [PETR](https://arxiv.org/abs/2203.05625) and [PETRv2](https://arxiv.org/abs/2206.01256), which might be interesting future extensions to BEVFusion.
