This forum is disabled, please visit https://forum.opencv.org
 | 1 | initial version |
no, you can't (or - you should not).
cascades degrade, if you increase the pose angle, if you have morethan ~ +=15°,you'll get random predictions.
while you can overcome it by training a frontal and a side view (and flip the image for the opposide side), this means you would need 3 seperate detection attempts.
also note, that the venerable cascade classifiers are "on the way out" (current opencv4 has no more tools to train those), and that you probably should use more modern cnn based face detectors. e.g. look at MTCNN, which is much faster than a single cascade already, andworks nice in [-90 .. 90] degree range. it does not make much sense to
| | 2 | No.2 Revision |
no, you can't (or - you should not).
cascades degrade, if you increase the pose angle, if you have morethan ~ +=15°,you'll get random predictions.
while you can overcome it by training a frontal and a side view (and flip the image for the opposide side), this means you would need 3 seperate detection attempts.
also note, that the venerable cascade classifiers are "on the way out" (current opencv4 has no more tools to train those), and that you probably should use more modern cnn based face detectors. e.g. look at MTCNN, which is much faster than a single cascade already, andworks nice in [-90 .. 90] degree range.
it does not make much sense to range.
