BEVFusion（mit）复现Ubuntu20.04终端配置

不解风水

已于 2024-04-12 15:37:00 修改

阅读量2.3k

点赞数 21

文章标签：深度学习 pytorch 人工智能

于 2023-12-12 16:04:54 首次发布

本文链接：https://blog.csdn.net/qq_60423839/article/details/134950050

版权

参考文章BEVFsion（mit）最强环境安装，部署复现

大致上参考这篇文章，但还是遇到了很多问题

可能原作者修改后会导致链接丢失，这里在文末将原作中重要步骤引用一遍。原作者在这里：

复现过程

原文：BEVFusion(mit)最强环境安装，部署复现

复现过程

首先查看cuda版本，原文要求11.3，这里使用的是11.1也没有问题

nvcc -V

因为服务器上有多个版本的cuda，参考此处切换linux系统下多个CUDA版本切换

创建switch-cuda.sh文件之后可以按以下指令切换

source switch-cuda.sh#查看版本
source switch-cuda.sh 11.1#选择自己要切换的版本

创建虚拟环境并安装torch

conda create -n bevfusion_mit python=3.8
 
#在安装torch的时候指定cuda版本，不容易出问题，cu113指cuda 11.3
pip install torch==1.10.0+cu113 torchvision==0.11.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html

因为本人使用的是cuda11.1，所以将cu113改为cu111

安装下列，记得先配置好镜像源再安装，不然会很慢。配置镜像源参考的这篇Anaconda配置镜像源

#安装mmcv的时候同样指定cuda和torch版本，cu113指cuda 11.3
pip install mmcv-full==1.4.0 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.10.0/index.html
 
pip install mmdet==2.20.0
 
conda install openmpi
 
conda install mpi4py
 
pip install Pillow==8.4.0
 
pip install tqdm
 
pip install torchpack
 
pip install nuscenes-devkit
 
pip install ninja
 
pip install numpy==1.19.5
 
pip install numba==0.48.0
 
pip install shapely==1.8.0

下载过程中会出现有关numpy版本问题，因为nuscenes-devkit需要大于1.22.0的版本，出现报错后直接再次装一遍numpy1.22.0

顺便一提，numba安装的很慢的话不要忘了在安装之前把镜像源配好。

下载BEVFusion

git clone https://github.com/mit-han-lab/bevfusion.git

这里可以直接去github上下载自己想要的代码

需要修改的是把mmdet3d/ops/spconv/src/indice_cuda.cu里面的4096都改成256

然后检查，要在虚拟环境中进入bevfusion文件夹目录下，开始编译

python setup.py develop

到这一步真的太不容易了，之前踩了好多坑导致没有编译成功，因为不知道运行成功的现象，也没有报错，没检查出来影响了后续步骤，正确运行的话差不多会出现这样的现象：

放在代码框里可能看得更清楚一些，这里前面部分出现了两个框，和原作者不太一样，但后续好像没有影响

(bevfusion) ysy@oaklong:~/bevfusion$ python setup.py develop
running develop
/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/setuptools/command/develop.py:40: EasyInstallDeprecationWarning: easy_install command is deprecated.
!!

        ********************************************************************************
        Please avoid running ``setup.py`` and ``easy_install``.
        Instead, use pypa/build, pypa/installer or other
        standards-based tools.

        See https://github.com/pypa/setuptools/issues/917 for details.
        ********************************************************************************

!!
  easy_install.initialize_options(self)
/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/setuptools/_distutils/cmd.py:66: SetuptoolsDeprecationWarning: setup.py install is deprecated.
!!

        ********************************************************************************
        Please avoid running ``setup.py`` directly.
        Instead, use pypa/build, pypa/installer or other
        standards-based tools.

        See https://blog.ganssle.io/articles/2021/10/setup-py-deprecated.html for details.
        ********************************************************************************

!!
  self.initialize_options()
running egg_info
writing mmdet3d.egg-info/PKG-INFO
writing dependency_links to mmdet3d.egg-info/dependency_links.txt
writing top-level names to mmdet3d.egg-info/top_level.txt
reading manifest file 'mmdet3d.egg-info/SOURCES.txt'
adding license file 'LICENSE'
writing manifest file 'mmdet3d.egg-info/SOURCES.txt'
running build_ext
/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/utils/cpp_extension.py:782: UserWarning: The detected CUDA version (11.1) has a minor version mismatch with the version that was used to compile PyTorch (11.3). Most likely this shouldn't be a problem.
  warnings.warn(CUDA_MISMATCH_WARN.format(cuda_str_version, torch.version.cuda))
building 'mmdet3d.ops.spconv.sparse_conv_ext' extension
Emitting ninja build file /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/build.ninja...
Compiling objects...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
[1/7] c++ -MMD -MF /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/reordering.o.d -pthread -B /home/ysy/.conda/envs/bevfusion/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/ysy/bevfusion/mmdet3d/ops/spconv/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/TH -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda-11.1/include -I/home/ysy/.conda/envs/bevfusion/include/python3.8 -c -c /home/ysy/bevfusion/mmdet3d/ops/spconv/src/reordering.cc -o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/reordering.o -w -std=c++14 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=sparse_conv_ext -D_GLIBCXX_USE_CXX11_ABI=0
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
[2/7] c++ -MMD -MF /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/maxpool.o.d -pthread -B /home/ysy/.conda/envs/bevfusion/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/ysy/bevfusion/mmdet3d/ops/spconv/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/TH -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda-11.1/include -I/home/ysy/.conda/envs/bevfusion/include/python3.8 -c -c /home/ysy/bevfusion/mmdet3d/ops/spconv/src/maxpool.cc -o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/maxpool.o -w -std=c++14 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=sparse_conv_ext -D_GLIBCXX_USE_CXX11_ABI=0
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
[3/7] c++ -MMD -MF /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/indice.o.d -pthread -B /home/ysy/.conda/envs/bevfusion/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/ysy/bevfusion/mmdet3d/ops/spconv/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/TH -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda-11.1/include -I/home/ysy/.conda/envs/bevfusion/include/python3.8 -c -c /home/ysy/bevfusion/mmdet3d/ops/spconv/src/indice.cc -o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/indice.o -w -std=c++14 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=sparse_conv_ext -D_GLIBCXX_USE_CXX11_ABI=0
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
[4/7] /usr/local/cuda-11.1/bin/nvcc  -DWITH_CUDA -I/home/ysy/bevfusion/mmdet3d/ops/spconv/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/TH -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda-11.1/include -I/home/ysy/.conda/envs/bevfusion/include/python3.8 -c -c /home/ysy/bevfusion/mmdet3d/ops/spconv/src/reordering_cuda.cu -o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/reordering_cuda.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options ''"'"'-fPIC'"'"'' -w -std=c++14 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_86,code=sm_86 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=sparse_conv_ext -D_GLIBCXX_USE_CXX11_ABI=0
[5/7] c++ -MMD -MF /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/all.o.d -pthread -B /home/ysy/.conda/envs/bevfusion/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/ysy/bevfusion/mmdet3d/ops/spconv/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/TH -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda-11.1/include -I/home/ysy/.conda/envs/bevfusion/include/python3.8 -c -c /home/ysy/bevfusion/mmdet3d/ops/spconv/src/all.cc -o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/all.o -w -std=c++14 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=sparse_conv_ext -D_GLIBCXX_USE_CXX11_ABI=0
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
[6/7] /usr/local/cuda-11.1/bin/nvcc  -DWITH_CUDA -I/home/ysy/bevfusion/mmdet3d/ops/spconv/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/TH -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda-11.1/include -I/home/ysy/.conda/envs/bevfusion/include/python3.8 -c -c /home/ysy/bevfusion/mmdet3d/ops/spconv/src/indice_cuda.cu -o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/indice_cuda.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options ''"'"'-fPIC'"'"'' -w -std=c++14 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_86,code=sm_86 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=sparse_conv_ext -D_GLIBCXX_USE_CXX11_ABI=0
[7/7] /usr/local/cuda-11.1/bin/nvcc  -DWITH_CUDA -I/home/ysy/bevfusion/mmdet3d/ops/spconv/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/TH -I/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda-11.1/include -I/home/ysy/.conda/envs/bevfusion/include/python3.8 -c -c /home/ysy/bevfusion/mmdet3d/ops/spconv/src/maxpool_cuda.cu -o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/maxpool_cuda.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options ''"'"'-fPIC'"'"'' -w -std=c++14 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_86,code=sm_86 -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=sparse_conv_ext -D_GLIBCXX_USE_CXX11_ABI=0
g++ -pthread -B /home/ysy/.conda/envs/bevfusion/compiler_compat -Wl,--sysroot=/ -pthread -shared -B /home/ysy/.conda/envs/bevfusion/compiler_compat -L/home/ysy/.conda/envs/bevfusion/lib -Wl,-rpath=/home/ysy/.conda/envs/bevfusion/lib -Wl,--no-as-needed -Wl,--sysroot=/ /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/all.o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/indice.o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/indice_cuda.o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/maxpool.o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/maxpool_cuda.o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/reordering.o /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/mmdet3d/ops/spconv/src/reordering_cuda.o -L/home/ysy/.conda/envs/bevfusion/lib/python3.8/site-packages/torch/lib -L/usr/local/cuda-11.1/lib64 -lc10 -ltorch -ltorch_cpu -ltorch_python -lcudart -lc10_cuda -ltorch_cuda_cu -ltorch_cuda_cpp -o build/lib.linux-x86_64-cpython-38/mmdet3d/ops/spconv/sparse_conv_ext.cpython-38-x86_64-linux-gnu.so
building 'mmdet3d.ops.bev_pool.bev_pool_ext' extension
Emitting ninja build file /home/ysy/bevfusion/build/temp.linux-x86_64-cpython-38/build.ninja...

后面还有很多，仅展示开头结尾即可

运行结束之后大概是这样

如果运行成功，就可以开始下载数据集了，还是参考文章开头的那篇来制作数据集

本人使用的是mini版本的，数据集下载好之后就可以运行数据转换的脚本

#同样的，如果是正常版本，运行：
python tools/create_data.py nuscenes --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes
 
#如果是mini版本，运行：
python tools/create_data.py  nuscenes --root-path ./data/nuscenes/ --version v1.0-mini --out-dir data/nuscenes/ --extra-tag nuscenes

在这个过程中出现的报错可以去检查前面的步骤，基本上setup.py运行成功这步就没什么问题

训练,官方给的是分布式训练：

torchpack dist-run -np 1 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

这里出现了报错RuntimeError: CUDA out of memory.

RuntimeError: CUDA out of memory. Tried to allocate 168.00 MiB (GPU 0; 23.70 GiB total 
capacity; 2.41 GiB already allocated; 96.50 MiB free; 2.43 GiB reserved in total by 
PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid 
fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

网上看到了很多解决方案说要修改batch_size,但是不太清楚这个如何修改，有的说在train.py 文件里，这里没有找到，最终的解决方案如下：

找到mmdet3d/apis/train.py

修改distributed为True

之前为distributed=False，修改为 distributed=True之后解决了。

运行结果如下

可视化

torchpack dist-run -np 1 python tools/visualize.py train_result/configs.yaml --mode gt --checkpoint train_result/latest.pth --bbox-score 0.5 --out-dir vis_result

可视化的结果拼图展示：

其他训练指令：

#训练只有相机的探测
torchpack dist-run -np 1 python tools/train.py configs/nuscenes/det/centerhead/lssfpn/camera/256x704/swint/default.yaml --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth
 
#训练只有相机的分割
torchpack dist-run -np 1 python tools/train.py configs/nuscenes/seg/camera-bev256d2.yaml --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth
 
#训练只有lidar的检测器
torchpack dist-run -np 1 python tools/train.py configs/nuscenes/det/transfusion/secfpn/lidar/voxelnet_0p075.yaml
 
#训练只有lidar的分割模型
torchpack dist-run -np 1 python tools/train.py configs/nuscenes/seg/lidar-centerpoint-bev128.yaml
 
#训练融合模型
torchpack dist-run -np 1 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

BEVFusion(mit)最强环境安装，部署复现

1.显卡驱动安装

如果你的电脑还没有安装显卡驱动，需要根据系统推荐的进行安装。查看是否有显卡驱动的指令是：

nvidia-smi

2.cuda安装

cuda的版本需要根据显卡来选择，运行nvidia-smi之后，可以看到显卡支持的最高cuda版本，那么我们安装的cuda版本要小于等于这个上限。但是又不能太低，有的显卡型号比较新，装了太老的cuda反而会出问题。我查到的是11.4，于是安装了cuda11.3，具体步骤如下：

（1）首先，进入下面的网址，点击对应版本的cuda。cuda选择官网

（2）在终端中运行生成的两个指令，分别是下载和安装cuda。下载到99%时出现段错误（核心已转储）。参考：安装CUDA段错误（核心已转储）解决方案

（3）安装时注意有几处需要选择，第一处选择continue，第二处输入accept，第三处用空格取消driver，并选择install。这样选择防止再次安装其他版本的显卡驱动导致出问题。

（4）配置环境变量

sudu gedit ~/.bashrc

在后面加上

    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
    export PATH=$PATH:/usr/local/cuda/bin
    export CUDA_HOME=$CUDA_HOME:/usr/local/cuda

保存之后，记得source一下

source ~/.bashrc

查看cuda是否安装成功

nvcc -V

3.环境配置

下面正式开始配环境啦。

（1）下载安装openmpi，我也不知道这个到底需不需要，安了没有副作用。

wget https://download.open-mpi.org/release/open-mpi/v4.1/openmpi-4.1.4.tar.gz
解压
cd openmpi-4.1.4
 
./configure –prefix=/usr/local/openmpi
 
make -j8
 
sudo make install
 
~/.bashrc文件里添加环境变量：
 
MPI_HOME=/usr/local/openmpi
OMPI_MCA_opal_cuda_support=true
export PATH=${MPI_HOME}/bin:$PATH
export LD_LIBRARY_PATH=${MPI_HOME}/lib:$LD_LIBRARY_PATH
export MANPATH=${MPI_HOME}/share/man:$MANPATH
 
测试安装是否成功
cd openmpi-x.x.x/examples
make
mpirun -np 4 hello_c

（2）创建虚拟环境并安装torch

conda create -n bevfusion_mit python=3.8
 
#在安装torch的时候指定cuda版本，不容易出问题，cu113指cuda 11.3
pip install torch==1.10.0+cu113 torchvision==0.11.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html

（3）安装下列...

#安装mmcv的时候同样指定cuda和torch版本，cu113指cuda 11.3
pip install mmcv-full==1.4.0 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.10.0/index.html
 
pip install mmdet==2.20.0
 
conda install openmpi
 
conda install mpi4py
 
pip install Pillow==8.4.0
 
pip install tqdm
 
pip install torchpack
 
pip install nuscenes-devkit
 
pip install ninja
 
pip install numpy==1.19.5
 
pip install numba==0.48.0
 
pip install shapely==1.8.0

（4）下载BEVFusion代码并做一些小修改

git clone https://github.com/mit-han-lab/bevfusion.git

1.把mmdet3d/ops/spconv/src/indice_cuda.cu里面的4096都改成256

               2.对于编译脚本 setup.py，需要把显卡算力改成自己对应的。在下面这个链接里查出你的显卡对应的算力，然后选择用哪一行：

        "-gencode=arch=compute_70,code=sm_70"

        "-gencode=arch=compute_75,code=sm_75"

        "-gencode=arch=compute_80,code=sm_80"

        "-gencode=arch=compute_86,code=sm_86"

CUDA GPUs - Compute Capability | NVIDIA Developer  查到3060显卡的算力为86.

3.开始编译

python setup.py develop

4.数据准备

（1）下载nuScenes数据集，如果有需要就下载完整版，学习代码可以下载mini

nuScenes官网https://www.nuscenes.org/nuscenes

(2) 把文件夹格式改成下面这个样子，注意nuscenes这个单词全部小写。

#正常版本，结构如下：
bevfusion-mit
├── tools
├── configs
├── data
│   ├── nuscenes
│   │   ├── maps
│   │   ├── samples
│   │   ├── sweeps
│   │   ├── lidarseg (optional)
│   │   ├── v1.0-test
|   |   ├── v1.0-trainval
 
 
#如果下载的是mini版本，结构如下：
bevfusion-mit
├── tools
├── configs
├── data
│   ├── nuscenes
│   │   ├── maps
│   │   ├── samples
│   │   ├── sweeps
│   │   ├── v1.0-mini

注意：还要下载Map expansion pack(v1.3) 然后解压到maps文件夹中。否则后面运行的时候会报错！

（3）接下来运行数据转换的脚本

#同样的，如果是正常版本，运行：
python tools/create_data.py nuscenes --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes
 
#如果是mini版本，运行：
python tools/create_data.py  nuscenes --root-path ./data/nuscenes/ --version v1.0-mini --out-dir data/nuscenes/ --extra-tag nuscenes

完成之后，文件夹是这样的，增加了几个.pkl文件和nuscenes_database。

#正常版本：
├── data
│   ├── nuscenes
│   │   ├── maps
│   │   ├── samples
│   │   ├── sweeps
│   │   ├── lidarseg (optional)
│   │   ├── v1.0-test
|   |   ├── v1.0-trainval
│   │   ├── nuscenes_database
│   │   ├── nuscenes_infos_train.pkl
│   │   ├── nuscenes_infos_val.pkl
│   │   ├── nuscenes_infos_test.pkl
│   │   ├── nuscenes_dbinfos_train.pkl
 
#mini版本
├── data
│   ├── nuscenes
│   │   ├── maps
│   │   ├── samples
│   │   ├── sweeps
│   │   ├── v1.0-mini
│   │   ├── nuscenes_database
│   │   ├── nuscenes_infos_train.pkl
│   │   ├── nuscenes_infos_val.pkl
│   │   ├── nuscenes_dbinfos_train.pkl

（4）下载预训练权重

5.终端训练与测试

（1）训练,官方给的是分布式训练：

torchpack dist-run -np 1 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

（2）测试：

torchpack dist-run -np 1 python tools/test.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml pretrained/bevfusion-det.pth --eval bbox