Implementing a ros package for interfacing FCN-8 with ROS. FCN-8 is described in (link) and the original code (not made by me) is available at (link).
(Tested on Ubuntu 16.04, ROS Kinetic, Python 2.7.12, opencv 3.2.0-dev, Cuda 8.0, cuDNN 6.0)
Follow the ROS installation guide for download and installation.
sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update
Get the newest driver with the “Software & Updates” application in the “Additional Drivers”-tab.
Software requires CUDA to be installed. Go to download page select your platform, download and follow instructions. In Select Target Platform i used Linux/x86_64/16.04/deb (local)
(You will need a nvidia user) Download cuDNN link. Extract zipped folder and copy to /usr/local/cuda-8.0
cd ~/Downloads/cuda (This is the directory of the extracted folder)
sudo cp ./include/* /usr/local/cuda-8.0/include/
sudo cp ./lib64/* /usr/local/cuda-8.0/lib64/
Install caffe dependencies
sudo apt-get install libboost-dev libprotobuf-dev protobuf-compiler libgflags-dev libgoogle-glog-dev libblas-dev libhdf5-serial-dev libopencv-dev libleveldb-dev liblmdb-dev libboost-all-dev libatlas-base-dev libsnappy-dev python-pip
pip install scipy
pip install numpy
pip install -U scikit-image
pip install protobuf
//apt-get install python-opencv
Get nvidias edition of caffe and git clone it to ~/Code/
cd && mkdir Code && cd Code
You may either select original caffe version from Berkeley (Caffe (Berkeley)) or nvidias edition (Caffe (nvidia)). git clone https://github.com/BVLC/caffe.git # Caffe (Berkeley) git clone https://github.com/NVIDIA/caffe.git # Caffe (nvidia)
In Makefile replace LIBRARIES line (around 183) with
LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_serial_hl hdf5_serial
Make a copy of Makefile.config.example
cd ~/Code/caffe
cp Makefile.config.example Makefile.config
Add /usr/include/hdf5/serial/ to INCLUDE_DIRS (around line 94-96)
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/
Include the following lines in the Makefile.config
USE_CUDNN := 1 (Only if GPU is used)
WITH_PYTHON_LAYER := 1
RESTART computer before building caffe sudo reboot now
Build caffe
cd ~/Code/caffe
make all -j8
make test -j8
make runtest -j8
make pycaffe
One-liner
make all -j8 && make test -j8 && make runtest -j8 && make pycaffe
Append the following lines to ~/.bashrc
# CAFFE
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export CAFFE_ROOT=~/Code/caffe # (IMPORTANT TO SET THIS AS THE INSTALL DIRECTORY OF CAFFE)
export PYTHONPATH=$CAFFE_ROOT/python:$PYTHONPATH
Git clone the package to your ros-workspace repository.
cd <your_workspace>/src
git clone https://github.com/PeteHeine/RosSemanticSegmentation
Get weights. (Prototxt and caffemodel is from here)
cd <your_workspace>/src/fcn8_ros/models/
wget http://dl.caffe.berkeleyvision.org/pascalcontext-fcn8s-heavy.caffemodel