Installation instructions for py-Faster-RCNN

py-faster-rcnn_setup.md

Install CUDA and Caffe dependencies

sudo apt-get update && sudo apt-get upgrade
sudo apt-get install -y opencl-headers build-essential protobuf-compiler \
    libprotoc-dev libboost-all-dev libleveldb-dev hdf5-tools libhdf5-serial-dev \
    libopencv-dev libopenblas-dev liblapack-dev libsnappy-dev libsnappy1 \
    cmake libstdc++6-4.8-dbg libgoogle-glog0 libgoogle-glog-dev \
    libgflags-dev liblmdb-dev git python-pip gfortran

Install CUDA

# kernel source dependecies for cuda drivers and cuda toolkit
sudo apt-get install -y linux-image-extra-`uname -r` linux-headers-`uname -r` linux-image-`uname -r`

# download & install CUDA 7.0 Ubuntu repository from NVidia's website
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.0-28_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1404_7.0-28_amd64.deb

# Update Ubuntu's package manager
sudo apt-get update && sudo apt-get upgrade

# Install package "cuda-7-0". It includes CUDA drivers, CUDA toolkit and other utilities
sudo apt-get install cuda-7-0

# check whether nvidia drivers are installed correctly. It would show details about  your GPU
nvidia-smi

# Configure configure ldconfig so that CUDA binaries are visible
sudo sh -c "sudo echo '/usr/local/cuda/lib64' > /etc/ld.so.conf.d/cuda.conf"
sudo ldconfig

Install cuDNN

To obtain cuDNN you have to register at NVidia's Accelerated Computing Developer Program After registration, you can download cuDNN from this link We have used cuDNN v4.

tar -xvf cudnn-7.0-linux-x64-v4.0-prod.tgz

# -P flag is important because it retains the symlinks
sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64/
sudo cp cuda/include/cudnn.h /usr/local/cuda/include

# Run ldconfig again
sudo ldconfig

Compile and install OpenCV

To install OpenCV use this automation script by Jay Rambhia. We will install OpenCV 2.4.10

# Open this script and define these 2 variables at the top of script.
# version=2.4.10 and
# downloadfile=opencv-2.4.10.zip

# At line 45, where `cmake` command is specified, add an extra flag which specifies GPU architecture:
cmake -D CMAKE_BUILD_TYPE=RELEASE -D CUDA_GENERATION=Kepler

# So the full `cmake` command would be:
cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local  -D CUDA_GENERATION=Kepler \
    -D WITH_TBB=ON -D BUILD_NEW_PYTHON_SUPPORT=ON -D WITH_V4L=ON -D INSTALL_C_EXAMPLES=ON \
    -D INSTALL_PYTHON_EXAMPLES=ON -D BUILD_EXAMPLES=ON -D WITH_QT=ON -D WITH_OPENGL=ON ..

Note: Don't install OpenCV 2.4.9. It has a bug due to which it gives compile error while building OpenCV CUDA modules.

Install Caffe & py-faster-rcnn

# Install py-faster-rcnn dependencies
sudo pip install Cython
sudo pip install easydict

git clone --recursive https://github.com/rbgirshick/py-faster-rcnn.git
cd py-faster-rcnn/lib
# IMPORTANT: Since we are using NVidia Kepler 520 GPU, which has compute capability of 3.0
# change line 135 in py-faster-rcnn/lib/setup.py
# from '-arch=sm_35' =====> '-arch=sm_30'

# compile py-faster-rcnn Cython modules
make

# Now we will compile Caffe
cd ../caffe-faster-rcnn
for req in $(cat requirements.txt); do sudo pip install -U $req; done

# prepare Makefile.config
cp Makefile.config.example Makefile.config
# make following changes
# Uncomment following lines
USE_CUDNN := 1
WITH_PYTHON_LAYER := 1
USE_PKG_CONFIG := 1
# Change BLAS and PYTHON_INCLUDE to following values
BLAS := open
PYTHON_INCLUDE := /usr/include/python2.7 \
                /usr/lib/python2.7/dist-packages/numpy/core/include \
                /usr/local/lib/python2.7/dist-packages/numpy/core/include/

# compile Caffe modules
make -j8 all

# Since this is a fork of original caffe. There is a small bug due to which
# compilation of test modules fails with following error message:
# test_smooth_L1_loss_layer.cpp:11:35: fatal error: caffe/vision_layers.hpp: No such file or directory
# Solution: https://github.com/rbgirshick/py-faster-rcnn/issues/155
# remove line 11 in src/caffe/test/test_smooth_L1_loss_layer.cpp where vision_layers.hpp is included
make -j8 test

# run tests to ensure all the tests are passed
make runtest
# build pycaffe
make pycaffe

Now that we have installed Caffe and py-faster-rcnn, we can follow demo and training instructions from py-faster-rcnn's Github repo

Thanks, this script helped me!