The reproduction project of the BEV model # BEVFusion, which includes some code annotation work.
Thanks for the BEVFusion authors！Paper | Code
- configs/
- data/nuscenes/
- maps/
- samples/
- sweeps/
- v1.0-test/
- v1.0-trainval/
- mmdet3d/
- tools/
- pretrained/
You can refer to the official configuration environment documentation. Official Git
BEVFusion's official introduction document is very comprehensive and detailed.
Or use the Conda env configuration file we provide.
conda env create -f bevfusion_env.yaml
After that, you should
python setup.py develop
python tools/create_data.py nuscenes --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes
./tools/download_pretrained.sh
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
torchpack dist-run -np 8 python tools/train.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth --load_from pretrained/lidar-only-det.pth
torchpack dist-run -np 8 python tools/test.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml pretrained/bevfusion-det.pth --eval bbox
| ID | Name | mAP | NDS | mATE | mASE | mAOE | mAVE | mAAE | Epochs | Data | Batch_size | GPUs | Train_time | Eval_time | Log_file |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | bevfusion cam+lidar | 0.6807 | 0.7107 | 0.2871 | 0.2531 | 0.3091 | 0.2632 | 0.1844 | 6 | All | 16, sample per gpu=2 | 8 x Nvidia Geforce 3090 | 12hours | 71.1s | runs/bevfusion_res_log/ |
Here we have made simple annotations on some key model files in Chinese, these annotations are based on "configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml" config.
You can find them in:
- mmdet3d\models\fusion_models\bevfusion.py