Since the optical flow example in Touch is just a fake and not a real optical flow / motion estimation implementation, it would be great if this Nvidia SDK could be supported in Touch.
They seem to cover different algorithms for optimization and it is supposed to work in realtime on 4K videos, which could be great for video interpolation or other tricks… Thank you 
@ch3 mentioned it and I looked a bit at the docs, great thing is it seems it’s using dedicated hardware on the gpu so it basically for free.
I guess it’s possible to implement it with a c++ TOP but an official implementation would be great.
+1 
+1 on this feature request.
I’ve compiled and run the examples from the SDK, but haven’t tried any c++ for Touch yet.
Great, let’s make some noise
I spend a few hours trying to bring the example code from the SDK into the TD cPlusPlus CUDA example… but I really don’t know what I am doing. haha
So many unknowns for me in there and also struggle to find any more docs on the API (other than this page), as well as CUDA examples for TD.
Thanks for taking the time, hope some C++ guru can help you!
If anyone needs this done, I’m available for freelance work to do it. I’ve learned some important stuff about CUDA lately https://derivative.ca/community-post/tensorflowpytorch-fast-c-tops-hire-me-freelance/62887 So I think I can get it done in a reasonable timeframe.