Hi, I am training the ViT KPRPE model, I checked the similarity change between the pretrained ViT KPRPE WebFace4M and ViT WebFace4M, the overall face similarity seems becomes lower for same identity.
And landmarks are generated by mobilenet face detector? it seems some wrong landmarks are generated for high angle side face.
Could you please share some ideas on this?
Hi, I am training the ViT KPRPE model, I checked the similarity change between the pretrained ViT KPRPE WebFace4M and ViT WebFace4M, the overall face similarity seems becomes lower for same identity.
And landmarks are generated by mobilenet face detector? it seems some wrong landmarks are generated for high angle side face.
Could you please share some ideas on this?