2x1=10

Note that now that BVLC’s caffe repos­i­to­ry direct­ly sup­ports com­pi­la­tion on Win­dows, this guide has become obso­lete.

When attempt­ing to check out Caffe after some wast­ed hours of get­ting Ten­sor­Flow to run on Win­dows (which sort-of works using Bash for Win­dows), I gave Caffe a try. How­ev­er, the last time I tried to get a huge sys­tem of nest­ed depen­den­cies using CMake to work on Win­dows, things only got worse with every addi­tion­al project. For Caffe, you’d need Boost, OpenCV, HDF and pro­to­buf as well as some Google log­ging and com­mand-line argu­ment things amongst oth­er stuff — and boy, make sure you got your link­ing, thread­ing and run­time right, and don’t even think about mix­ing dif­fer­ent C++ or Boost fla­vors.

So there’s this Win­dows fork of Caffe man­aged by Microsoft which appar­ent­ly only requires you to con­fig­ure the CommonSettings.props file to your lik­ing and then com­pile. The nice part is, all ref­er­ences are pulled using NuGet. The bad part is, the pack­ages are all MSVC 1800, i.e. Visu­al Stu­dio 2013. You’ll be able to find new­er vari­ants of Boost, but that’s about it.

If you’re real­ly try­ing to use Visu­al Stu­dio 2015, fix the nuget.config to have the repos­i­to­ry path fixed like described in this Stack­Over­flow answer:

<?xml version="1.0" encoding="utf-8"?>
<configuration>
  <packageSources>
    <add key="nuget.org" value="https://www.nuget.org/api/v2/" />
  </packageSources>
  <add key="repositoryPath" value="..\..\NugetPackages" />
</configuration>

Also note that the path is most like­ly out­side your build tree … but that’s how it is pro­vid­ed by the main­tain­er.
I wast­ed anoth­er day try­ing to get Caffe to work in Visu­al Stu­dio 2015 by select­ing the 2013 toolk­it but, in the end, it came down to the fol­low­ing error:

NuGet error: unknown command 'overlay' 

This overlay stuff appar­ent­ly is some NuGet 2 mag­ic that didn’t sur­vive the dark ages, so it won’t work with NuGet 3, which I have in my path and which is also used inter­nal­ly by VS 2015.
There are a cou­ple of ques­tions regard­ing that error, like this one on the OpenCV Answers site, issues on GitHub (with half-bro­ken answers since the CoApp tool­ing doesn’t real­ly work exist any­more), etc., but there’s no obvi­ous solu­tion — at least if you’re using a mod­ern Win­dows devel­op­ment envi­ron­ment — that is, one that’s using NuGet 3 instead of NuGet 2.6.

You can fix most of the prob­lems by just upgrad­ing OpenCV 2 to OpenCV 3. For that, kick out the old NuGet pack­ages and ref­er­ence the opencvde­fault one as shown in this video. But even if you do that, it’ll fail on the glog 0.3.3 NuGet pack­age which still uses over­lays.

As it turns out though, the NuGet pack­age for glog 0.3.3 already con­tains a NuGet bina­ry at glog.0.3.3.0/build/native/private/nuget.exe, but since it’s not in the path, it won’t use it. Copy that to the package’s root (where NuGet-Overlay.cmd can be found and you’re set.

For the rest, make sure you’re using Python 2.7 though, as the Boost libraries are requir­ing that; also install NumPy.
Using e.g. Ana­con­da Python, you’d want

conda create -n caffe python=2.7 numpy

Which will leave you with the task to find python27_d.lib … or just build Release.