Opencv 3.3, Ubuntu 16.04, Cuda 9.1 Installation Quick Checks

gistfile1.txt

Download NVIDIA and CUDA drivers from here (.run files)
https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=runfilelocal
http://www.nvidia.com/content/DriverDownload-March2009/confirmation.php?url=/XFree86/Linux-x86_64/390.25/NVIDIA-Linux-x86_64-390.25.run&lang=us&type=TITAN

disable nouveau
disable XServer
rm /tmp .X* files
sudo init 3
then run cuda 

Install opencv
https://medium.com/@debugvn/installing-opencv-3-3-0-on-ubuntu-16-04-lts-7db376f93961

opencv cuda error:
Change these files accordingly: OpenCVDetectCUDA.cmake, FindCUDA.cmake as per the link below
https://stackoverflow.com/questions/46584000/cmake-error-variables-are-set-to-notfound

In, 
opencv-3.3.0\modules\cudev\include\opencv2\cudev\common.hpp
Add #include <cuda_fp16.h>



In OpenCVDetectCUDA.cmake, change :
CUDA_ARCH_BIN : 2.0 3.0 3.5 3.7 5.0 5.2 6.0 6.1

to :
CUDA_ARCH_BIN : 6.0 6.1


-----------
Tensorflow
1.6 works with cuda 9.1 and above
1.4 works with cuda 8.0

Opencv_3.3_Ubuntu16_Cuda_9.1_Installation.txt

Download NVIDIA and CUDA drivers from here (.run files)
https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=runfilelocal
http://www.nvidia.com/content/DriverDownload-March2009/confirmation.php?url=/XFree86/Linux-x86_64/390.25/NVIDIA-Linux-x86_64-390.25.run&lang=us&type=TITAN

disable nouveau
disable XServer
rm /tmp .X* filesuname -r
sudo update-initramfs -u [update initramfs]

sudo init 3
then run cuda 

sudo killall /usr/bin/X

Build gcc 5.3.1 (when a pre-existing gcc is present)
cd gcc 5.3.1
sudo apt-get install build-essential
./contrib/download_prerequisites 
mkdir build && cd build
../configure --enable-checking=release --enable-languages=c,c++ --disable-multilib

ubuntu-drivers devices [ this will print all available drivers]


CUDA Installation
sudo update-grub && uname -r [to check the kernel being used] If it's wrong, then log out, and enter advanced system setup, select 'linux-headers-4.4.0-128-generic' at login. We are going to install cuda on this kernel image
sudo service lightdm stop
Download NVIDIA driver and run it [NVIDIA-Linux-x86_64-390.48.run] 
Download cuda_*.run and chmod it [cuda_9.1.85_387.26_linux.run]
uname -r : will give tyou the kernel images available. We are going to use 4.4.0.128
provide path to kernel image while installing cuda
sudo ./cuda_9.1.85_387.26_linux.run 
Select 'YES' for all (including nvidia-xconfig, OpenGL..)
This will replace nvidia 390 by 387
export PATH=/usr/local/cuda-9.1/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-9.1/lib64{LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
echo -e "/usr/local/cuda-9.1/lib64" > /etc/ld.so.conf.d/cuda.conf
sudo ldconfig
sudo reboot now
which nvcc [this should display /usr/local/cuda-9.1/bin/nvcc]
nvcc -V [should print 9.1]
nvidia-smi [387.26]
sudo nvidia-xconfig

If the output is negative for nouveau, then all is well with your installation.
lsmod | grep nouveau 

Ubuntu login loop after driver installation:
chown ppillai:ppillai .Xauthority
sudo mokutil --disable-validation
This will ask for pwd: Type verkada31

-----
Restart XServer
sudo nvidia-xconfig [this will initialize the NVIDIA X Server :) :) Almost there..]
You can check the X-Server conf file contents contents : cat /etc/X11/xorg.conf

Also set linux-headers-4.4.0-128-generic as the default grub login (instead of manually selecting it everytime)
sudo 
apt install -y -f linux-headers-$(uname -r)
sudo cp /etc/default/grub /etc/default/grub.bak
sudo -H gedit /etc/default/grub
GRUB_DEFAULT="Advanced options for Ubuntu>Ubuntu, with Linux 3.13.0-53-generic" 
sudo update-grub

----
Latest guide
----
sudo apt-get install gedit-plugins

sudo add-apt-repository ppa:gnome-terminator
sudo apt-get update
sudo apt-get install terminator

https://github.com/heethesh/Install-TensorFlow-OpenCV-GPU-Ubuntu-17.10
building bazel using this:
tensorflow/tools/pip_package:build_pip_package --action_env="LD_LIBRARY_PATH=${LD_LIBRARY_PATH}"

Install opencv
https://medium.com/@debugvn/installing-opencv-3-3-0-on-ubuntu-16-04-lts-7db376f93961

opencv cuda error:
Change these files accordingly: OpenCVDetectCUDA.cmake, FindCUDA.cmake as per the link below
https://stackoverflow.com/questions/46584000/cmake-error-variables-are-set-to-notfound

In, 
opencv-3.3.0\modules\cudev\include\opencv2\cudev\common.hpp
Add #include <cuda_fp16.h>



In OpenCVDetectCUDA.cmake, change :
CUDA_ARCH_BIN : 2.0 3.0 3.5 3.7 5.0 5.2 6.0 6.1

to :
CUDA_ARCH_BIN : 6.0 6.1


-----------
Tensorflow
1.6 works with cuda 9.1 and above
1.4 works with cuda 8.0