Caffe Installation In Linux Mint 18 with CUDA, cuDNN

Introduction

This is only a personal log which contains the problems I encountered when I was installing the Caffe Deep Learning framework. My computer contains a Nvidia GTX980 Graphic Card, so it is suitable for me to install Caffe with CUDA and cuDNN supports.

I also added python3 support in the configuration file.

The installation guides of the Caffe official website are pretty well written, but sometimes it is still a tedious experience to installing it. So I decide to share the problems that I have encountered and how I solved them.

For your convenience, I have attached my Makefile.config in this repo and you should modify it to match your own env if you want to use it. :)

make:protoc: command not found

Solution:

sudo apt-get install protobuf-c-compiler protobuf-compiler

fatal error: gflags/gflags.h: file or directory not found

Solution:

sudo apt-get install libgflags-dev

fatal error: glog/logging.h: file or directory not found

Solution:

sudo apt-get install libgoogle-glog-dev

fatal error: hdf5.h: file or directory not found

Solution:

sudo apt-get install libhdf5-*

or if you use zsh use the following:

sudo apt-get install libhdf5-\*

Wait， it is still not done yet. You have to add the hdf5 include files to the "Makefile.config" file:

# find this line:
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include

# modify it to:
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/

fatal error: lmdb.h: file or directory not found

Solution:

 sudo apt-get install liblmdb-dev

make: *** [.build_release/lib/libcaffe.so.1.0.0-rc3] Error 1

The complete outputs maybe like the following:

 AR -o .build_release/lib/libcaffe.a
 LD -o .build_release/lib/libcaffe.so.1.0.0-rc3
 /usr/bin/ld: cannot find -lboost_python3
 collect2: error: ld returned 1 exit status
 Makefile:563: recipe for target '.build_release/lib/libcaffe.so.1.0.0-rc3' failed
 make: *** [.build_release/lib/libcaffe.so.1.0.0-rc3] Error 1

Solution: Still you have to modify the config file:

 # find the following lines:
 INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial

 LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib
   
 # modify them to the following:
 # if your machine is 64 bit:
   INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial

   LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu /usr/lib/x86_64-linux-gnu/hdf5/serial
 # if your machine is 32 bit:
   INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial

   LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/i386-linux-gnu /usr/lib/i386-linux-gnu/hdf5/serial

LD ...atlas or blas: not found

In the linking stage the compiler can not find atlas or blas library files:

Solution:

  sudo apt-get install libopenblas-dev

Then modify the config file:

# find this:
BLAS := atlas

# modify it to:
BLAS := open

LD: ...cuDNN lib: file not found:

Solution: I have installed the cuDNN using the .deb file. It is also need the cudnn-8.0-linux-x64-v?.?.tgz file and extract it to the "/usr/local/cuda" directory and merge those files into it. (one include file to the include directory and lib64 directory to the lib64 directory)