1. 代码下载

git clone https://github.com/fundamentalvision/BEVFormer.git

2. 环境配置

使用conda创建环境

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

进入环境

conda activate open-mmlab

允许使用pip安装依赖库

export PIP_REQUIRE_VIRTUALENV=false

安装pytorch和torchvision

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

Install gcc>=5 in conda env (optional).

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

安装mmcv-full

pip install addict
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

安装mmdet和mmseg

pip install mmdet==2.14.0
pip install mmsegmentation==0.14.1

Install mmdet3d from source code.

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

Install Detectron2 and 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'

3. 准备数据

3.1. 下载数据

下载地址 https://www.nuscenes.org/download

从这里点击下载

3.1.1. 下载Mini数据

需要下载Mini, Test和can_bus，得到如下文件

v1.0-mini.tgz

v1.0-test_blobs.tgz

v1.0-test_meta.tgz

can_bus.zip

3.1.2. 下载全量数据

需要下载TrainVal, Test和can_bus，得到如下文件

v1.0-trainval01_blobs.tgz
v1.0-trainval02_blobs.tgz
v1.0-trainval03_blobs.tgz
v1.0-trainval04_blobs.tgz
v1.0-trainval05_blobs.tgz
v1.0-trainval06_blobs.tgz
v1.0-trainval07_blobs.tgz
v1.0-trainval08_blobs.tgz
v1.0-trainval09_blobs.tgz
v1.0-trainval10_blobs.tgz
v1.0-trainval_meta.tgz

v1.0-test_blobs.tgz

v1.0-test_meta.tgz

can_bus.zip

3.2. 组织数据

3.2.1. 组织Mini数据

把Mini的压缩包都放在同一个目录，例如名为v1.0-mini，然后解压，解压的时候会有一些目录合并。

把v1.0-mini改名为v1.0-trainval，这样得到的目录是

maps

samples

sweeps

v1.0-test

v1.0-trainval

can_bus

3.2.2. 组织全量数据

把全量的压缩包都放在同一个目录，例如名为v1.0-full，然后解压，解压的时候会有一些目录合并。

maps 

samples 

sweeps

v1.0-test

v1.0-trainval

can_bus

3.3. 预处理数据

python tools/create_data.py nuscenes --root-path [your path]/v1.0-mini/ --out-dir [your path]/v1.0-mini/ --extra-tag nuscenes --version v1.0 --canbus [your path]

4. 训练

打开文件projects/configs/bevformer/bevformer_base.py，修改data_root.

./tools/dist_train.sh ./projects/configs/bevformer/bevformer_base.py 8

5. 评测

./tools/dist_test.sh ./projects/configs/bevformer/bevformer_base.py ./path/to/ckpts.pth 8

参考文献

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

https://github.com/fundamentalvision/BEVFormer/blob/master/docs/getting_started.md