1.https://github.com/fundamentalvision/BEVFormer下载解压
2.cd BEVFormer,下载解压 https://github.com/open-mmlab/mmdetection3d/tree/v0.17.1 (注意版本0.17.1)
3.pytorch版本
conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge
4.cd mmdetection3d-0.17.1,运行pip install -v -e .和pip install -r requirements(各别库的版本适配)
5.mmcv有问题,官网下载https://download.openmmlab.com/mmcv/dist/cu113/torch1.10.0/index.html
6.detectron2有问题,官网下载 https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html
7.在tran.py添加D:\code\BEVFormer-master\my_bevformer_tiny.py配置文件
8.数据集生成方式可见官网
9.配置文件只修改
data_root='./BEVFormer-master/data/nuscenes/',
ann_file='./BEVFormer-master/data/nuscenes/nuscenes_infos_temporal_train.pkl',