Since middle of Aug. ,we have bought a new server with 2 TitanX GPUs,I was studying about how to play with caffe.
1.
install NVIDIA drivers and CUDA 7.5
2.
install PYTHON (ipython notebook)
sudo apt-get install -y ipython-notebook pandoc
3.
install MATLAB
4.
I change the BLAS edition (in Makefile.config) into OpenBLAS.
sudo apt-get install libgsl0-dev -y sudo apt-get install libblas3 libblas3gf liblapack-dev libsnappy-dev libatlas-dev libopenblas-dev -y
GOOGLE (gflags, protobuf)
sudo apt-get install protobuf-c-compiler protobuf-compiler -y
sudo apt-get install libgflags-dev -y
sudo apt-get install libprotobuf-dev -y
We should download glogs ourselves and compile it.
wget https://google-glog.googlecode.com/files/glog-0.3.3.tar.gz
tar -vxf glog-0.3.3.tar.gz
./configure; make -j4; sudo make install
BOOST
sudo apt-get install libboost1.55-all-dev -y
OpenCV
sudo apt-get install libopencv-* -y
STORAGE
Data blobs are transfered via DB format. In dev branch, we should use more types of them.
sudo apt-get install libhdf5-dev libleveldb-dev liblmdb-dev -y
ps:
PYTHON
Python depends on the tools that drawing the net or detect the object. If you wanna use these utilities, install them.
sudo apt-get install python-protobuf python-numpy python-skimage python-pydot python-pandas -y
error while loading shared libraries: libcudart.so.7.5: cannot open shared object file: No such file or directory
sudo gedit /etc/ld.so.conf
add:
/usr/local/cuda-7.5/lib64/
save and:
sudo ldconfig