Hi,
- I've seen that in the new FastCV release 1-7-0, there is a new procedure for Histogram of Oriented Gradient (HOG) features extraction called fcvExtractHOGu16(). First of all, thanks a lot for adding this procedure.
- A few questions though:
- 1. there are a lot of ways to implement HOG. The most common are the original one by Dallal & Triggs and another one which seems to be really fast by Piotr Dollar. Which implementation is used in the FastCV library? or Is the FastCV implementation comparable to the one provided by Matlab or OpenCV?
- 2. the input are the "strenght" and "orientation". Do you mean strenght == magnitude = sqrt(Gx^2 + Gy^2) and orientation== phase= atan2(Gx,Gy)?
- 3. Is the lenght of the hogVector calculated the same way as Matlab do it? - http://www.mathworks.com/help/vision/ref/extracthogfeatures.html
- 4. What is the difference between FASTCV_HOG_NORM_REGULAR , FASTCV_HOG_NORM_RENORMALIZATION and FASTCV_HOG_NORM_FHOG?
- 5. Is the bolckStep parametter the overlap between blocks? binSize, the number of orientations in the Histogram?...
- 6. Can you point me to a paper, a link, descrbing in more details the FastCV implementation of HOG.
Thanks a lot for your help.
Ariel