It’s about .net wrapper for DLib.
We have started it for a project, currently couple face detectors and video tracker are running - cuda accelerated (whats possible), - and runs realtime.
Adding features one by one is relatively easy, but making a full wrapper is changeling, of course. What would be really interesting is to setup a custom trainer for detectors - they saying it gives a result even on 5 images, and have a shiny example of it.

Here it is:


  • Cuda 10.0;
  • DlibDotNet.CUDA100 nuget:
    nuget install DlibDotNet.CUDA100 -Version
    or easier: open VL.DlibDotNet.vl and on the missing dependency click Install.

looks tasty

Any updates on this?

I would be interested and could contribute (with work).

Hi @ravazquez, there is not much progress yet, but I will publish it by the weekend.

1 Like

Later I should make a nuget for this, but for now build is included to the repo.

And when anybody has an idea of what to do with this mess of different types - like rectangles, points and vectors, - I would appreciate a hint.

You mean a Rectangle coming from a dll that’s not compatible with gamma’s built-in Rectangle ? You have to make your own conversion operations that converts from the lib’s Rectangle to gamma’s one, usually by splitting one and joining it to the other (and of course vice versa).

1 Like

Hi @sebescudie, this part is clear for me. But I’m wondering if there is a way to avoid producing hundreds of types which are actually the same but from different libraries: there are two types of Rectangle in Dlib, one in gamma and one in OpenCV… And when we are go to the vectors - it can go totaly crazy.