windows 11编译caffe 使用Anaconda3+python 3.5.6+cmake 3.40+Visual Studio 2015

1、安装 Visual Studio 2015
https://download.microsoft.com/download/5/d/1/5d1ec81e-bc59-448f-9ab6-27636d5cc18a/vs2015.3.com_chs.iso

2、安装Cmake 把cmake/bin 配置到PATH中

3、安装Anaconda3

4、下载caffe
https://github.com/BVLC/caffe/tree/windows?tab=readme-ov-file

5、开始编译
5.1 先下载 libraries_v140_x64_py35_1.1.0.tar.bz2 到
C:/Users/***/.caffe/dependencies/download/

5.2 执行编译

conda create -n caffe python=3.5
conda activate caffe 

cd \caffe-windows
scripts\build_win.cmd

报错： Found cuDNN: ver. ??? found

-- CUDA detected: 12.5
-- Found cuDNN: ver. ??? found (include: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.5/include, library: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.5/lib/x64/cudnn.lib)
CMake Error at cmake/Cuda.cmake:236 (message):
  cuDNN version >3 is required.
Call Stack (most recent call first):
  cmake/Cuda.cmake:264 (detect_cuDNN)
  cmake/Dependencies.cmake:92 (include)
  CMakeLists.txt:80 (include)

修改cmake\Cuda.cmake 文件第 179行和210行，因为新版本CUDA不同

问题：ImportError: No module named ‘numpy’

pip install numpy

问题 ：nvcc fatal : Unsupported gpu architecture ‘compute_20’

5.3 生成动态库

双击 caffe-windows\build\Caffe.sln

5.4 构建python接口

conda config --add channels conda-forge
conda config --add channels willyd

5.5 安装软件包

conda install --yes cmake ninja numpy scipy protobuf==3.1.0 six scikit-image pyyaml pydotplus graphviz

或者将caffe-windows\python\caffe 文件夹复制到python 的 site_packages