环境：Linux, Python=3.6.8, CUDA = 11.1, pytorch = 1.9.0, mmdet3d = 0.17.1
创建虚拟环境：conda create -n petr python=3.6.8 conda activate petr
安装torch等相关库
Install MMCVpip install mmcv-full==1.4.0 -f https://download.openmmlab.com/mmcv/dist/cu112/torch1.9.0/index.html
Install MMDetection
git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
git checkout v2.24.1
pip install -r requirements/build.txt
python3 setup.py develop
cd ..
Install MMSegmentation. pip install mmsegmentation==0.20.2
Install MMDetection3D
git clone https://github.com/open-mmlab/mmdetection3d.git
cd mmdetection3d
git checkout v0.17.1
pip install -r requirements/build.txt
python3 setup.py develop
cd ..
如果上面运行python3 setup.py develop的时候有什么库没安装上就用pip install xxx -i https://pypi.tuna.tsinghua.edu.cn/simple来下载
如果显示： libgeos_c.so缺失
安装geosconda install -c conda-forge geos=3.9.1
Install PETR
git clone https://github.com/megvii-research/PETR.git
cd PETR
mkdir ckpts #放权重
mkdir data #放数据集
ln -s {mmdetection3d_path} ./mmdetection3d #进行软连接
ln -s {nuscenes_path} ./data/nuscenes
下载nuScenes数据集：需注册并下载nuScenes数据集，放置到data/nuscenes目录，我用的是mini进行测试。
pip install nuscenes-devkit==1.1.10
生成预处理数据： python tools/create_data.py nuscenes --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes --version v1.0-mini应该会生成 nuscenes_infos_val.pkl文件
在官网中下载预训练权重：
可以下载官方训练好的权重选择正确的config文件进行推理：epoch_24.pth
修改config文件的data路径，例如：·data_root = ‘xxxx/PETR/data/nuscenes/’·
模型训练
tools/dist_train.sh projects/configs/petr/petr_vovnet_gridmask_p4_1600x640.py 2 --work-dir work_dirs/petr_vovnet_gridmask_p4_1600x640/
模型测试
tools/dist_test.sh work_dirs/petr_vovnet_gridmask_p4_1600x640/petr_vovnet_gridmask_p4_1600x640.py ckpts/epoch_24_1600x640.pth 8 --eval bbox