Caffe (0) Ubuntu 17.10 install CUDA, cnDNN, OpenCV

1. Download CUDA9.0 from the official site:

The address of the official site is https://developer.nvidia.com/cuda-90-download-archive

Select and download the runfile:

2. Blacklist nouveau:

Modify blacklist.conf:

$ sudo vim /etc/modprobe.d/blacklist.conf

Add these to the blacklist:

# this one might not be required for x86 32 bit users.
blacklist amd76x_edac 

blacklist vga16fb
blacklist nouveau
blacklist rivafb
blacklist nvidiafb
blacklist rivatv

Update initramfs disk:

$ sudo update-initramfs -u

3. Stop gdm3 and install the graphic driver:

$ sudo /etc/init.d/gdm3 stop
$ sudo init 3

Install nvidia-384 before installing CUDA. Don't use other version !!

$ runlevel # check runlevel is 3

$ sudo add-apt-repository ppa:graphics-drivers/ppa
$ sudo apt update
$ sudo apt install nvidia-384 nvidia-384-dev
$ sudo apt-get install g++ freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libglu1-mesa libglu1-mesa-dev g++-6 gcc-6

$ sync
$ sudo init 6 # reboot the system

Make sure that the driver has been installed correctly.

$ nvidia-smi

4. Install CUDA:

Use g++-6 and gcc-6 as default compiler:

$ cd /usr/bin
$ sudo rm g++ gcc
$ sudo ln -s g++-6 g++
$ sudo ln -s gcc-6 gcc

Install CUDA:

$ cd ~/Downloads
$ sudo chmod +x cuda_9.0.176_384.81_linux.run
$ sudo ./cuda_9.0.176_384.81_linux.run

DON'T install the driver !!

Edit the enviroment variables:

$ vim ~/.bashrc

#[start-20180602-ben(cuda)-add]#
export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
#[end-20180602-ben(cuda)-add]#

Build CUDA examples

$ cd ~/NVIDIA_CUDA-9.0_Samples/
$ make -j4

Run the example:

$ cd ~/NVIDIA_CUDA-9.0_Samples/bin/x86_64/linux/release
$ ./smokeParticles

5. Install cuDNN:

Go to the official web site. Login account and download cuDNN. Select "Download cuDNNv7.1.4 (May 16, 2018), for CUDA 9.0"

Don't select the deb file.

Unzip the file and install it:

$ tar -zxvf *.tgz
$ cd cuda

$ sudo cp -P include/cudnn.h /usr/include
$ sudo cp -P lib64/libcudnn* /usr/lib/x86_64-linux-gnu/
$ sudo chmod a+r /usr/lib/x86_64-linux-gnu/libcudnn*

6. Download the OpenCV and OpenCV contrib:

https://opencv.org/releases.html

https://github.com/opencv/opencv_contrib/releases

Make a folder and unzip the files:

Install cmake-gui:

$ sudo apt-get update
$ sudo install cmake-gui

Run cmake-gui and set the path:

Push the Configure button and select Unix Makefiles:

Make sure that cuda is selected:

Make sure that python3 is selected:

Push the Generate button.

Install OpenCV:

$ cd ~/OpenCV/opencv-3.4.1/build
$ make -j4
$ sudo make install

7. Install Caffe:

Clone Caffe project https://github.com/BVLC/caffe

Go to Caffe dir and modify Makefile.config.example:

$ cp Makefile.config.example Makefile.config
$ vim Makefile.config 

Follow the comment and modify the options in Makefile.config.

Build Caffe

$ make -j 4

Issue: hdf5.h not found. Solution https://github.com/NVIDIA/DIGITS/issues/156

cd /usr/lib/x86_64-linux-gnu
sudo ln -s libhdf5_serial.so.8.0.2 libhdf5.so
sudo ln -s libhdf5_serial_hl.so.8.0.2 libhdf5_hl.so

Modify INCLUDE_DIRS and LIBRARY_DIRS in Makefile.config.

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/hdf5/serial/

Issue: error: token ""__CUDACC_VER__ is no longer supported. Solution: https://github.com/BVLC/caffe/issues/5994.

For me, I just comments /usr/local/cuda/include/crt/common_functions.h line 64:
#define __CUDACC_VER__ "__CUDACC_VER__ is no longer supported. Use __CUDACC_VER_MAJOR__, __CUDACC_VER_MINOR__, and __CUDACC_VER_BUILD__ instead."

Test Caffe:

$ make runtest

8. Install PyCaffe:

Build PyCaffe:

$ make pycaffe

To import the caffe Python module after completing the installation, add the module directory to your $PYTHONPATH by export PYTHONPATH=/path/to/caffe/python:$PYTHONPATH

Reference:

1. https://github.com/BVLC/caffe/issues/4843
2. https://blog.csdn.net/DonatelloBZero/article/details/51304162