Contrastive Loss Coupled with Occlusion Aware Training Aids in Face Recognition from Low Quality Group Photos

In scenarios without restrictions, recognizing faces from a group photo is still a difficult task. There are two major challenges: (i) the faces can be partially occluded, (ii) due to low resolution facial features may appear similar among multiple faces. State of the art methods are found to suffer in the presence of these challenges. In this paper we propose a method to mitigate this limitation. We artificially occlude training images to make the face recognition models be aware of occlusion. Secondly, we put a contrastive loss in a cohort to avoid the face recognition model being confused between similar looking faces due to low quality. In multiple datasets we have shown that state-of-the-art face recognition models improve their performance with the proposed training mechanism.