caffe-cpu | Implementing the Wasserstein Loss Layer | Machine Learning library

Problem solved: The error came up because caffe build wasn't done successfully. I recommend not to go up with the sudo apt install caffe-cpu command (Which is mentioned in the official caffe installation guide for Ubuntu); because it will end up in the error as above. It's better to install from the source.

Let me give step by step guidance to install caffe successfully in Ubuntu 18.04 LTS:

1] sudo apt-get install -y --no-install-recommends libboost-all-dev

2] sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libboost-all-dev libhdf5-serial-dev \ libgflags-dev libgoogle-glog-dev liblmdb-dev protobuf-compiler

3] git clone https://github.com/BVLC/caffe cd caffe cp Makefile.config.example Makefile.config

4] sudo pip install scikit-image protobuf cd python for req in $(cat requirements.txt); do sudo pip install $req; done

5] Modify the Makefile.config file: Uncomment the line CPU_ONLY := 1, and the line OPENCV_VERSION := 3.

6] Find LIBRARIES line in Makefile and change it to as follows:

LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_hl hdf5 \ opencv_core opencv_highgui opencv_imgproc opencv_imgcodecs

7] make all

Now you could get some error like this:

CXX src/caffe/net.cpp src/caffe/net.cpp:8:18: fatal error: hdf5.h: No such file or directory compilation terminated. Makefile:575: recipe for target '.build_release/src/caffe/net.o' failed make: *** [.build_release/src/caffe/net.o] Error 1

To solve this error follow step 8.

8] install libhdf5-dev open Makefile.config, locate line containing LIBRARY_DIRS and append /usr/lib /x86_64-linux-gnu/hdf5/serial locate INCLUDE_DIRS and append /usr/include/hdf5/serial/ (per this SO answer) rerun make all

9] make test

10] make runtest

11] make pycaffe

Now you could get some error like this:

CXX/LD -o python/caffe/_caffe.so python/caffe/_caffe.cpp python/caffe/_caffe.cpp:10:31: fatal error: numpy/arrayobject.h: No such file or directory compilation terminated. Makefile:501: recipe for target 'python/caffe/_caffe.so' failed make: *** [python/caffe/_caffe.so] Error 1

To solve this error follow step 12.

12] Find PYTHON_INCLUDE line in Makefile.config and do the changes as follows:

I was able to get it working following these steps,

1. Get caffe source form here (https://github.com/BVLC/caffe.git)

2. Install CUDA if you need GPS support (https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1804)

3. Install CUDNN, if you need GPS support (https://developer.nvidia.com/rdp/cudnn-download)

4. Replace existing Makefile.config with this (https://gist.github.com/GPrathap/1f9d184c55779509860b8bf92cea416d) Here I have configured for cuda 9.2. If you have a different version please search 9.2 and change the version which you have instaled. Also, please recheck all the paths which are declared in Makefile.config.

5. You may type make all followed by make test

6. make distribute for creating the final lib, an include directory of caffe which can be found in (caffe/distribute).

7. If you are using CMake type project add where to find caffe as bellow,

When you install caffe on ubuntu using sudo apt install caffe-cpu, it compiles the bindings for python 3 only (_caffe.cpython-36m-x86_64-linux-gnu.so), which is located at /usr/lib/python3/dist-packages/caffe/. So the short answer is to use python 3 instead.

The long answer is to compile caffe with python 2 bindings from source.

I tried on my end to install caffe-cpu, and it seems that the file /etc/apt/sources.list doesn't have the needed repositories to install it, in the datalab instance. To workaround this issue, I used the following commands, in a created notebook:

The Ubuntu packages for caffe install the caffe executable in /usr/bin. The particular file you want is in caffe-tools-cpu, not caffe-cpu as such.

From https://packages.ubuntu.com/bionic/amd64/caffe-tools-cpu/filelist:

/usr/bin/ld: cannot find -lhdf5_hl

-lhdf5_hl means libhdf5_hl.so