Is there any deep CNN to extract HOG/HOOF descripters?


l’m looking for a learned CNN to extract HOG (histogram of oriented gradient) and HOOF (histogram of optical flow) features. I’m wondering if there is any CNN which do that.

Thank you

Just our of curiosity, what is your use case?
Is the vanilla calculation of HOG/HOOF too bad/slow?
Would you like to improve it, e.g. like optical flow?

Hi @ptrblck. Yes definitely l would like to improve it like optical flow . l would like to get a deep representation of HOG/HOOF

Hi @ptrblck,

Any clue ?

Here is my explicit question : CNN for getting and representing superpixels