serengil/retinaface

detect_faces image preprocessing?

utility-aagrawal opened this issue · 2 comments

Hi,

I see that images are upscaled before running them through the model in detect_faces(). Why is that so? My understanding is that retinaface expects input image of 640 x 640 but that's not what images are upscaled to? Can you please explain the reasoning behind the upscaling?

Thanks!

Hi,

As I added comments into the source code, that logic is copied from another repo. TBH, I do not know the reason behind it.

No problem, Thanks for your response! Let me see if they can answer it.