博客

环境：Ubuntu20.04，cudnn8.1.1.33-1+cuda11.2，opencv4.2

caffe编译安装

  https://github.com/BVLC/caffe

注：caffe只支持到cudnn7.x，我鼓捣了一阵（直接把cudnn相关文件换掉，解决了一部分问题），放弃使用cudnn。https://github.com/BVLC/caffe/issues/6970

（1）安装依赖包

sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler

sudo apt install --no-install-recommends libboost-all-dev

sudo apt install libopenblas-dev liblapack-dev libatlas-base-dev

sudo apt install libgflags-dev libgoogle-glog-dev liblmdb-dev

sudo apt install git cmake build-essential

（2）修改 Makefile.config

cd caffe

cp Makefile.config.example Makefile.config

vim Makefile.config

    详情

  修改代码：
#USE_OPENCV := 0
#OPENCV_VERSION := 3
#WITH_PYTHON_LAYER := 1
  为：
USE_OPENCV := 1
OPENCV_VERSION := 4
WITH_PYTHON_LAYER := 1

  修改代码：
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib
  为：
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial /usr/local/include/opencv4
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu /usr/lib/x86_64-linux-gnu/hdf5/serial /usr/local/bin
  # 最后是自己的opencv的include和bin路径，解决fatal error: opencv2/core/core.hpp问题

  修改代码：
PYTHON_INCLUDE := /usr/include/python2.7 \
                /usr/lib/python/dist-packages/numpy/core/include
  为：
PYTHON_INCLUDE := /usr/include/python3.8 \
                /usr/lib/python3/dist-packages/numpy/core/include
  # 根据自己的路径修改

  修改代码：
CUDA_ARCH := -gencode arch=compute_20,code=sm_20 \
		-gencode arch=compute_20,code=sm_21 \
		-gencode arch=compute_30,code=sm_30 \
		-gencode arch=compute_35,code=sm_35 \
		-gencode arch=compute_50,code=sm_50 \
		-gencode arch=compute_52,code=sm_52 \
		-gencode arch=compute_60,code=sm_60 \
		-gencode arch=compute_61,code=sm_61 \
		-gencode arch=compute_61,code=compute_61
  为：
CUDA_ARCH := -gencode arch=compute_35,code=sm_35 \
		-gencode arch=compute_50,code=sm_50 \
		-gencode arch=compute_52,code=sm_52 \
		-gencode arch=compute_60,code=sm_60 \
		-gencode arch=compute_61,code=sm_61 \
		-gencode arch=compute_61,code=compute_61
  # 根据自己的架构和版本修改，可参考https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/

（3）修改Makefile

vim Makefile

  修改代码：
NVCCFLAGS +=-ccbin=$(CXX) -Xcompiler-fPIC $(COMMON_FLAGS)
  为：
NVCCFLAGS += -D_FORCE_INLINES -ccbin=$(CXX) -Xcompiler -fPIC $(COMMON_FLAGS)

  修改代码：
LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_hl hdf5
  为：
LIBRARIES += glog gflags protobuf boost_system boost_filesystem hdf5_hl hdf5 m opencv_core opencv_highgui opencv_imgproc opencv_imgcodecs
  # 解决undefined reference to cv::imread(cv::String const&, int)...,可参考https://github.com/BVLC/caffe/issues/2348

  修改代码：
PYTHON_LIBRARIES ?= boost_python python2.7
  为：
PYTHON_LIBRARIES ?= boost_python38 python3.8
  # 根据自己的版本修改

（4）支持opencv4

修改src/caffe/util/io.cpp，解决error: ‘CV_LOAD_IMAGE_COLOR’ was not declared in this scope

vim src/caffe/util/io.cpp

  修改代码：
cv_img = cv::imread(image.first, CV_LOAD_IMAGE_COLOR);
  为：
cv_img = cv::imread(image.first,cv::IMREAD_COLOR);
  # 需要修改两处

修改src/caffe/layers/window_data_layer.cpp，解决error: ‘CV_LOAD_IMAGE_GRAYSCALE’ was not declared in this scope…

vim src/caffe/layers/window_data_layer.cpp

  修改代码：
int cv_read_flag = (is_color ? CV_LOAD_IMAGE_COLOR :
    CV_LOAD_IMAGE_GRAYSCALE);
  为：
int cv_read_flag = (is_color ? cv::IMREAD_COLOR :
    cv::IMREAD_GRAYSCALE);

在src/caffe/data_transformer.cpp中添加opencv头文件，解决.build失败

vim src/caffe/data_transformer.cpp

  修改代码：
#ifdef USE_OPENCV
#include <opencv2/core/core.hpp>
#endif  // USE_OPENCV
  为：
#ifdef USE_OPENCV
#include <opencv2/core/core.hpp>
#include <opencv2/core/mat.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/highgui/highgui_c.h>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/imgproc/imgproc_c.h>
#include <opencv2/opencv.hpp>
#endif  // USE_OPENCV

在 include/caffe/common.hpp中增加opencv4支持（第70行后）

vim include/caffe/common.hpp

  新增代码：
// Supporting OpenCV4
#if (CV_MAJOR_VERSION == 4)
#define CV_LOAD_IMAGE_COLOR cv::IMREAD_COLOR
#define CV_LOAD_IMAGE_GRAYSCALE cv::IMREAD_GRAYSCALE
#endif

  以下为原代码（用于定位）：
// See PR #1236
namespace cv { class Mat; }

（5）编译

make all -j8

make pycaffe -j8

若import caffe 报错 Traceback (most recent call last): File “<stdin>”, line 1, in <module> ModuleNotFoundError: No module named ‘caffe’

sudo echo export PYTHONPATH="~/caffe/python" >> ~/.bashrc

source ~/.bashrc

参考文章：https://blog.csdn.net/yhaolpz/article/details/71375762