yubangji123
commented
2 years ago On way, by using our network, you can find the indexes of the occlusion parts within the feature vectors, if you want to remove the occlusion related features. Another way is to use any up-to-date FR methods, as far as I know, many recent FR methods can generate occlusion robust features to perform strong enough face verification.
If i want to verify whether the two faces with occlusion belong to one identity，do I need any extra operation on their 320-D features？do I need to discard the features affected by the occlusion？If yes，how to do？