Caffe Unet installation - anan-dad/quadnet GitHub Wiki

Unet is a powerful deep learning architecture in tasks like semantic segamentation. We don't need to install cuda, but we need to make sure we add them into the env. Cuda is available at "/misc/softwares/cuda" at LMB, make sure with your tutor that you have access to the folders. Choose the cuda version you need and add them into env:

$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/misc/software/cuda/cudnn-9.2.88-cudnn-7.3.1/lib64
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/misc/software/cuda/cudnn-9.2.88-cudnn-7.3.1
$ export PATH=$PATH:/misc/software/cuda-cuda-9.2.88-cudnn-7.3.1

Next, we need a specific commit d1208dbf313698de9ef70b3362c89cfddb51c520 of caffe for building caffe_unet.

$ git clone https://github.com/BVLC/caffe.git
$ cd caffe
$ git checkout d1208dbf313698de9ef70b3362c89cfddb51c520

Next, we need a patch file to build caffe_unet from source, right click here to save the patch file. Then execute:

$ mv /home/path_to_the_patch_file/caffe_unet.patch ..
$ git apply ../caffe_unet.patch
$ mkdir build
$ cd build
$ ccmake ..
$ make -j8
$ make install
$ make runtest

If you encounter any problems, please click here for the official instructions page. For more details about caffe_unet, please refer to this website.