BEVDet

1.环境配置

参考：官方文档
参考：BEVDet复现

1. 基本配置

需求：ubuntu20.04，python-3.8, torch1.13.0, cuda-11.7, cudnn-8.6, conda
未安装可以参考

2. 基本依赖

apt-get install -y vim libsm6 libxext6 libxrender-dev libgl1-mesa-glx git wget libssl-dev libopencv-dev libspdlog-dev

3. 拉取源码

# 1 下载源码   目前默认是dev3.1
git clone https://github.com/HuangJunJie2017/BEVDet.git

# 2 查看代码版本号 显示dev3.1版本
git checkout

# 3 如果要切换版本号，如dev2.1，输入
git checkout dev2.1

4. 配置bevdet的python虚拟环境

# 1 创建虚拟环境
conda create -n bevdet python=3.8

# 2 激活python环境
conda activate bevdet

# 3 虚拟环境bevdet中安装torch
pip install torch==1.13.0+cu117 torchvision==0.14.0+cu117 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu117

# 4 虚拟环境bevdet中安装openlib相关库
pip install mmcv-full==1.5.3 onnxruntime-gpu==1.8.1 mmdet==2.25.1 mmsegmentation==0.25.0

# 5 进入BEVDet工程目录,安装mmdet3d
pip install -e -v .

# 6 安装其他依赖 numpy==1.23.4 setuptools==58.2.0等
pip install pycuda lyft_dataset_sdk networkx==2.2 numba==0.53.0 numpy==1.23.4 nuscenes-devkit plyfile scikit-image tensorboard trimesh==2.35.39 setuptools==58.2.0 yapf==0.40.1
 

2. 解决问题

2.1 pip装包编译错误

参考：https://blog.csdn.net/h904798869/article/details/131719404
参考：https://blog.csdn.net/weixin_62075168/article/details/129758163

问题：

    src/cpp/cuda.hpp:14:10: fatal error: cuda.h: 没有那个文件或目录
       #include <cuda.h>
                ^~~~~~~~
      compilation terminated.
      /tmp/pip-build-env-lze7ntp3/overlay/lib/python3.8/site-packages/setuptools/command/build_py.py:207: _Warning: Package 'pycuda.cuda' is absent from the `pa                                                                              ckages` configuration.
      !!

              ********************************************************************************
              ############################
              # Package would be ignored #
              ############################
              Python recognizes 'pycuda.cuda' as an importable package[^1],
              but it is absent from setuptools' `packages` configuration.

              This leads to an ambiguous overall configuration. If you want to distribute this
              package, please make sure that 'pycuda.cuda' is explicitly added
              to the `packages` configuration field.

              Alternatively, you can also rely on setuptools' discovery methods
              (for example by using `find_namespace_packages(...)`/`find_namespace:`
              instead of `find_packages(...)`/`find:`).

              You can read more about "package discovery" on setuptools documentation page:

              - https://setuptools.pypa.io/en/latest/userguide/package_discovery.html

              If you don't want 'pycuda.cuda' to be distributed and are
              already explicitly excluding 'pycuda.cuda' via
              `find_namespace_packages(...)/find_namespace` or `find_packages(...)/find`,
              you can try to use `exclude_package_data`, or `include-package-data=False` in
              combination with a more fine grained `package-data` configuration.

              You can read more about "package data files" on setuptools documentation page:

              - https://setuptools.pypa.io/en/latest/userguide/datafiles.html


              [^1]: For Python, any directory (with suitable naming) can be imported,
                    even if it does not contain any `.py` files.
                    On the other hand, currently there is no concept of package data
                    directory, all directories are treated like packages.
              ********************************************************************************

      !!
        check.warn(importable)
      error: command '/usr/bin/gcc' failed with exit code 1
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
  ERROR: Failed building wheel for pycuda
Failed to build pycuda
ERROR: Could not build wheels for pycuda, which is required to install pyproject.toml-based projects

解决思路：

required to install pyproject.toml-based projects这个问题主要出现在C++构建工具未安装

Windows平台解决思路Linux平台解决思路

在Windows平台安装比较繁琐，首先要下载Vs,一般来说可以直接安装Visual Studio，但是占用空间大，只安装构建也能解决问题

在其中标题栏选择下载一栏，并在下载的搜索框键入：build tools，并回车进入下载界面

下载如图的工具，下载完成以后是一个ISO结尾的光盘映射文件，注意下载的时候将类型改为DVD

双击运行该光盘映射文件，点击exe文件，一路进行安装

按图片勾选的工具进行安装，即可解决

下载依赖

sudo apt install g++ make

解决完问题接下来就继续pip安装即可。

2.2 MMCV工具安装

问题：

(bevdet) root@wahaha-System-Product-Name:/media/wahaha/222/BEVDet/bev/BEVDet2/BEVDet# python tools/create_data_bevdet.py                                   /root/miniconda3/envs/bevdet/lib/python3.8/site-packages/mmcv/__init__.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv. See https://github.com/open-mmlab/mmcv/blob/master/docs/en/compatibility.md for more details.
  warnings.warn(
Traceback (most recent call last):
  File "tools/create_data_bevdet.py", line 9, in <module>
    from tools.data_converter import nuscenes_converter as nuscenes_converter
  File "/media/wahaha/222/BEVDet/bev/BEVDet2/BEVDet/tools/data_converter/nuscenes_converter.py", line 14, in <module>
    from mmdet3d.core.bbox import points_cam2img
  File "/media/wahaha/222/BEVDet/bev/BEVDet2/BEVDet/mmdet3d/__init__.py", line 4, in <module>
    import mmdet
  File "/root/miniconda3/envs/bevdet/lib/python3.8/site-packages/mmdet/__init__.py", line 24, in <module>
    assert (mmcv_version >= digit_version(mmcv_minimum_version)
AssertionError: MMCV==1.7.0 is used but incompatible. Please install mmcv>=1.3.17, <=1.6.0.

在运行脚本创建数据集时，按照官方配置安装的MMCV工具提示版本不兼容，需要安装其他版本的，但是安装完所推荐的版本，又会提示不兼容，安装其他的版本，反复横跳

解决思路

自己修改MMCV工具版本需求范围
打开MMCV工具目录，找到__init__.py文件，找到如下的范围
把这三个文件的__init__.py文件中的MMCV_MIN,MMCV_MAX改成包含你现在拥有的版本，忽略版本范围


再次运行脚本，发现脚本运行成功

自动生成了训练集与验证集

2.3 其他问题···

3.数据集的创建

参考：https://blog.csdn.net/h904798869/article/details/133363474
参考：https://blog.csdn.net/h904798869/article/details/130317240?spm=1001.2014.3001.5502

3.1 下载数据集

nuscenes下载地址
下载细节参考 FastBEV

由于我们使用的mini数据集进行测试，而源码中需要full数据集，将v1.0-mini复制一份并命名为v1.0-trainval``./data
文件结构如下：

data
  └──nuscenes
    ├── gts
    ├── maps
    ├── samples
    ├── sweeps
    ├── v1.0-mini
    └── v1.0-trainval  # 复制的文件夹

3.2 生成bevdet数据集

运行tools/create_data_bevdet.py生成数据集

python tools/create_data_bevdet.py

结果如图：

4.测试

首先通过测试来实现论文效果，直接使用官方训练好的权重，详情请见，官方权重

python tools/test.py ./configs/bevdet/bevdet-r50.py ckpts/bevdet-r50.pth --eval mAP

5.可视化

5.1 生成json

# 运行test.py 必须--out", "--eval", "--format-only", "--show" or "--show-dir至少跟一个
# json文件生成需要增加 --eval-options参数 jsonfile_prefix=test_dirs
# 实在搞不清楚请看test.py的源码，看如何加载参数即可

# 1 直接测试
python tools/test.py ./configs/bevdet/bevdet-r50.py ckpts/bevdet-r50.pth --format-only

# 2 测试保存json文件
python tools/test.py ./configs/bevdet/bevdet-r50.py ckpts/bevdet-r50.pth --format-only --eval-options jsonfile_prefix=test_dirs
# 保留json位于目录test_dirs下

# 3 直接生成保存为pkl格式
python tools/test.py ./configs/bevdet/bevdet-r50.py ckpts/bevdet-r50.pth --out=./test_dirs/out.pkl

5.2 json文件转可视化

python tools/analysis_tools/vis.py ./work_dir/results_nusc.json

生成视频./vis/vis.mp4文件, 如下：

6.训练

7. 鸣谢

最感谢的是LHL小可爱的支持和努力，解决了无数的问题，完成了配置测试