List your company
ResearchFormerly WizWand

Interaction Recognition on ImageNet 1k (3 objects)

90Interaction Recognition AccuracyMETAGAME
24.4841.4958.575.51May 7, 2026
Updated 5mo ago

Evaluation results

MethodLinks
METAGAMEModel (Size)=SigLIP-2 (ViT-B/32), Base Method=MaskSigLIP2026.0590
METAGAMEModel (Size)=CLIP (ViT-B/16), Explanation Basis=Attention2026.0589
METAGAMEModel (Size)=CLIP (ViT-B/16), Explanation Basis=MaskCLIP2026.0589
METAGAMEModel (Size)=SigLIP-2 (ViT-B/32), Base Method=Grad-ESigLIP2026.0589
METAGAMEModel (Size)=SigLIP-2 (ViT-L/16), Base Method=MaskSigLIP2026.0588
FIxLIPModel (Size)=SigLIP-2 (ViT-B/32)2026.0587
METAGAMEModel (Size)=MetaCLIP-2 (ViT-H/14), Explanation Basis=Attention2026.0585
METAGAMEModel (Size)=SigLIP-2 (ViT-L/16), Base Method=Grad-ESigLIP2026.0584
METAGAMEModel (Size)=CLIP (ViT-B/16), Explanation Basis=Grad-ECLIP2026.0582
FIxLIPModel (Size)=SigLIP-2 (ViT-L/16)2026.0581
FIxLIPModel (Size)=CLIP (ViT-B/16)2026.0578
METAGAMEModel (Size)=MetaCLIP-2 (ViT-H/14), Explanation Basis=Grad-ECLIP2026.0578
METAGAMEModel (Size)=SigLIP-2 (ViT-B/32), Base Method=Attention2026.0578
METAGAMEModel (Size)=SigLIP-2 (ViT-L/16), Base Method=Attention2026.0576
FIxLIPModel (Size)=MetaCLIP-2 (ViT-H/14)2026.0573
MaskSigLIPModel (Size)=SigLIP-2 (ViT-L/16)2026.0554
MaskSigLIPModel (Size)=SigLIP-2 (ViT-B/32)2026.0542
Grad-ESigLIPModel (Size)=SigLIP-2 (ViT-B/32)2026.0533
AttentionModel (Size)=CLIP (ViT-B/16)2026.0532
AttentionModel (Size)=MetaCLIP-2 (ViT-H/14)2026.0532
Grad-ECLIPModel (Size)=CLIP (ViT-B/16)2026.0531
Grad-ECLIPModel (Size)=MetaCLIP-2 (ViT-H/14)2026.0531
Grad-ESigLIPModel (Size)=SigLIP-2 (ViT-L/16)2026.0531
AttentionModel (Size)=SigLIP-2 (ViT-B/32)2026.0530
AttentionModel (Size)=SigLIP-2 (ViT-L/16)2026.0529
MaskCLIPModel (Size)=CLIP (ViT-B/16)2026.0527