MINTIME-Multi-Identity-size-iNvariant-TIMEsformer-for-Video-Deepfake-Detection icon indicating copy to clipboard operation
MINTIME-Multi-Identity-size-iNvariant-TIMEsformer-for-Video-Deepfake-Detection copied to clipboard

Published 20 hours ago • verified publisher icon davide-coccomini

Detection rate；research direction

Open liyanou224 opened this issue 1 year ago • 1 comments
trafficstars

Hello! May I ask: 1) what is the approximate detection rate（frames per second） of the system you designed？ 2) why are there still many papers that have not considered multiple identities at present?

liyanou224 avatar Dec 01 '23 01:12 liyanou224

Hello, we consider 1 frame per second for our experiments but there it can be set as you desire. In general, the multi-identity case is pretty specific and very few datasets have this situation. Recently, a new dataset has been released that try to give us more multi-identity videos. https://openaccess.thecvf.com/content/CVPR2023/papers/Narayan_DF-Platter_Multi-Face_Heterogeneous_Deepfake_Dataset_CVPR_2023_paper.pdf

davide-coccomini avatar Dec 06 '23 12:12 davide-coccomini