【BEV】BEVFormer——利用时空 Transformer 从多相机图像中学习鸟瞰图表示

论文

论文的题目是：*《BEVFormer: Learning Bird’s-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformers》*

复现

环境安装

官方教程：https://github.com/fundamentalvision/BEVFormer/blob/master/docs/install.md

1. 创建conda环境

conda create -n open-mmlab python=3.8 -y
conda activate open-mmlab

2. 安装pytorch

pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
# Recommended torch>=1.9

3. 安装gcc6

conda install -c omgarcia gcc-6 # gcc-6.2

4. 安装mmcv-full、mmdet、mmseg

pip install mmcv-full==1.4.0
#  pip install mmcv-full==1.4.0 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html
pip install mmdet==2.14.0
pip install mmsegmentation==0.14.1

5. 安装mmdet3d

   git clone https://github.com/open-mmlab/mmdetection3d.git
   cd mmdetection3d
   git checkout v0.17.1 # Other versions may not be compatible.
   pip install -e .

6. 安装detectron2和Timm

   pip install einops fvcore seaborn iopath==0.1.9 timm==0.6.13  typing-extensions==4.5.0 pylint ipython==8.12  numpy==1.19.5 matplotlib==3.5.2 numba==0.48.0 pandas==1.4.4 scikit-image==0.19.3 setuptools==59.5.0
   python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'

7. 下载BEVFormer

   git clone https://github.com/fundamentalvision/BEVFormer.git
   cd bevformer
   mkdir ckpts
   cd ckpts & wget https://github.com/zhiqi-li/storage/releases/download/v1.0/r101_dcn_fcos3d_pretrain.pth