Caffe

Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by Berkeley AI Research (BAIR)/The Berkeley Vision and Learning Center (BVLC) and community contributors.

Check out the project site for all the details like

• DIY Deep Learning for Vision with Caffe
• Tutorial Documentation
• BAIR reference models and the community model zoo
• Installation instructions

and step-by-step examples.

Custom distributions

• Intel Caffe (Optimized for CPU and support for multi-node), in particular Intel® Xeon processors.
• OpenCL Caffe e.g. for AMD or Intel devices.
• Windows Caffe

Community

Please join the caffe-users group or gitter chat to ask questions and talk about methods and models. Framework development discussions and thorough bug reports are collected on Issues.

Happy brewing!

License and Citation

Caffe is released under the BSD 2-Clause license. The BAIR/BVLC reference models are released for unrestricted use.

Please cite Caffe in your publications if it helps your research:

@article{jia2014caffe,
  Author = {Jia, Yangqing and Shelhamer, Evan and Donahue, Jeff and Karayev, Sergey and Long, Jonathan and Girshick, Ross and Guadarrama, Sergio and Darrell, Trevor},
  Journal = {arXiv preprint arXiv:1408.5093},
  Title = {Caffe: Convolutional Architecture for Fast Feature Embedding},
  Year = {2014}
}

---

快速安装指南

系统 Centos7 踩坑记录 基于源码master9b89154的改动:

1. 支持opencv-4.3, 安装脚本; .zshrc
2. 支持cuda-11.1, cudnn-8
3. 支持python38, boost_python38, numpy-1.19.4 protobuf-3.6.0
4. 支持gcc-9.1.0, 安装脚本
5. 支持cmake-3.20.1, 安装脚本

安装C++依赖

新建一个conda环境，python==3.8； 或者老环境里一定要删除protobuf

conda uninstall protobuf
conda uninstall libprotobuf
pip uninstall protobuf
which protobuf

conda install -c conda-forge glog
conda install -c conda-forge zlib
conda install -c conda-forge openblas
conda install -c conda-forge lmdb
conda install -c conda-forge leveldb
conda install -c conda-forge boost
conda install -c conda-forge hdf5

版本对照表

    # packages in environment at /home/zhangwenting/anaconda3/envs/beta:
    #
    # Name                    Version                   Build  Channel
    glog                      0.4.0                h49b9bf7_3    conda-forge
    zlib                      1.2.11            h516909a_1010    conda-forge
    openblas                  0.3.12          pthreads_h43bd3aa_1    conda-forge
    lmdb                      0.9.24               h516909a_0    conda-forge
    leveldb                   1.22                 h7cfaab3_1    conda-forge
    boost                     1.74.0           py38hf6732f7_2    conda-forge
    hdf5                      1.12.0          nompi_h54c07f9_102    conda-forge

gcc-9.1.0 安装脚本 cmake-3.20.1 安装脚本 opencv-4.3 安装脚本 protobuf-3.6.0 源码编译:

wget https://github.com/google/protobuf/releases/download/v3.6.0/protobuf-3.6.0.tar.gz
tar -zxvf protobuf-3.6.0.tar.gz 
cd protobuf-3.6.0/ 
./configure --prefix=/your/install/path 
make 
make install

自动搜索配置

cd caffe
mkdir build
cd build
cmake ..

手动配置路径

cd build
ccmake ..

完整参考

这一步自动生成 build/Makefile

编译

cd build
make -j"$(nproc)" 1> make-5-6.log 2> make-5-6.error 

日志

安装测试

cd build
make install 1> install-5-6.log 2> install-5-6.error

安装日志

---

cd build
make runtest 1> runtest-5-6.log 2> runtest-5-6.error 

测试日志

编译安装python接口

• 编译

cd build
make pycaffe 1> pycaffe-5-6.log 2> pycaffe-5-6.error 

编译日志

---

• 安装

cd build
make pycaffe install
-- Up-to-date: /home/zhangwenting/software/caffe/build/install/python/caffe/_caffe.so

检查pycaffe的_caffe.so对应依赖是否完整

ldd $HOME/software/caffe/build/install/python/caffe/_caffe.so > ldd_caffe.so.log

依赖路径

---

• 版本发布 caffe的python接口也依赖于C代码，所以发布的pycaffe版本里头文件库文件都包含

cp -r ~/software/caffe/build/install ~/software/caffe-1.0.0-linux-x86_64
cp -r ~/software/caffe/build/install/python ~/software/pycaffe-1.0.0

• pycaffe依赖安装

cd ~/software/pycaffe-1.0.0
pip install -r requirements.txt
ll ~/software/pycaffe-1.0.0

python中引入caffe模块

• 途径一：环境变量中加入

export PYTHONPATH=~/software/pycaffe-1.0.0:$PYTHONPATH
python -c "import caffe"

• 途径二：每次导入caffe包的时候，加入搜索路径

import sys
sys.path.insert(0, "/home/zhangwenting/software/caffe-1.0.0-linux-x86_64/python")
import caffe

• 途径三：caffe包放入lib/python3.8/site-packages

cd site-packages
ln -s ~/software/pycaffe-cu11-1.0.0/caffe .
python -c "import caffe"

Release

For all: caffe-1.0.0-linux-x86_64 For python: pycaffe-1.0.0

• Support: CentOS Linux release 7.2.1511 CUDA-11.1 CuDNN-8
• Support: gcc 9.1.0 c++11 cmake 3.20.1
• Support: opencv-4.3
• Support: python3.8 numpy 1.19.4 protobuf-3.6

pycaffe环境移植指南

C++依赖

lld pycaffe-1.0.0/_caffe.so 看缺少哪些动态库，就安装哪些动态库

conda install -c confa-forge 库名称 如果有 sudo 权限就 yum install apt-get 找不到的就自己编译一下，包管理工具的好处：解决了依赖之间的版本对应关系，避免踩坑

conda install -c conda-forge glog=0.4.0 
conda install -c conda-forge lmdb=0.9.24
conda install -c conda-forge openblas
conda install -c conda-forge leveldb=1.18
conda install -c conda-forge boost=1.74
conda install -c conda-forge hdf5=1.12

并且，将这些库添加到$LD_LIBRARY_PATH 再动态装载：lld pycaffe-1.0.0/_caffe.so

如果依赖是安装于conda的虚拟环境下的，激活python虚拟环境时，执行默认的sh conda.sh仅更新$PATH，不更新$LD_LIBRARY_PATH，而将虚拟环境下的动态库路径写入rc的$LD_LIBRARY_PATH，非常不优雅。所以，需要修改conda.sh脚本，改成动态更新$LD_LIBRARY_PATH(激活虚拟环境时，更新$LD_LIBRARY_PATH，退出虚拟环境时，恢复$LD_LIBRARY_PATH)

python依赖

conda create -n caffe python==3.7
source activate caffe
# Download source and tar -xzvf
wget -c https://github.com/chenkangyang/caffe/releases/download/v1.0/pycaffe-1.0.0.tar.gz 
tar -zxvf pycaffe-1.0.0.tar.gz
cd pycaffe-1.0.0
pip install -r requirements
cd /your/path/to/site-packages
ln -s pycaffe-1.0.0/caffe .
python -c "import caffe"

最省事的一键安装(机器软硬件版本较新的话不推荐)

直接安装conda 的 intel镜像里最新的caffe-1.1.6-py36_intel_1包 (2019-09-06 02:26)

conda create -n caffe python==3.6
source activate caffe
conda install -c intel caffe