Detecting and recognizing faces seems to be difficult since I get very few faces detected but persons are detected just about every time.
I would like an option to do multiple face detection passes with different models to improve the odds of finding faces, like HOG, CNN, Caffe and whatnot instead of just picking one.
For saving unknown faces you could just loop over all selected models and save the crops but some more logic might be needed for returning coordinates to avoid intersections. Perhaps just something simple like the first hit, or some math to ensure no intersections/duplicates.
Detecting and recognizing faces seems to be difficult since I get very few faces detected but persons are detected just about every time.
I would like an option to do multiple face detection passes with different models to improve the odds of finding faces, like HOG, CNN, Caffe and whatnot instead of just picking one.
For saving unknown faces you could just loop over all selected models and save the crops but some more logic might be needed for returning coordinates to avoid intersections. Perhaps just something simple like the first hit, or some math to ensure no intersections/duplicates.
Thanks!