Visual Task Adaptation on VTAB
62.66VTAB Mean AccuracyTuneCLIP
Evaluation results
| Method | Links | |
|---|---|---|
| TuneCLIPBase Model=SigLIP ViT-B/162026.01 | 62.66 | |
| BaselineBase Model=SigLIP ViT-B/162026.01 | 60.39 | |
| FastCLIPBase Model=LAION ViT-B/322026.01 | 55.55 | |
| TuneCLIPBase Model=LAION ViT-B/322026.01 | 55.09 | |
| BaselineBase Model=LAION ViT-B/322026.01 | 55.08 | |
| OpenCLIPBase Model=LAION ViT-B/322026.01 | 54.68 | |
| TuneCLIPBase Model=OpenAI ViT-B/162026.01 | 54.44 | |
| BaselineBase Model=OpenAI ViT-B/162026.01 | 53.8 | |
| TuneCLIPBase Model=OpenAI ViT-B/322026.01 | 53.5 | |
| OpenCLIPBase Model=SigLIP ViT-B/162026.01 | 53.38 | |
| OpenCLIPBase Model=OpenAI ViT-B/162026.01 | 52.46 | |
| BaselineBase Model=OpenAI ViT-B/322026.01 | 51.81 | |
| FastCLIPBase Model=OpenAI ViT-B/162026.01 | 50.44 | |
| FastCLIPBase Model=SigLIP ViT-B/162026.01 | 49.93 | |
| OpenCLIPBase Model=OpenAI ViT-B/322026.01 | 49.23 | |
| FastCLIPBase Model=OpenAI ViT-B/322026.01 | 48.51 | |
| CE-CLIP2026.04 | 39.33 | |
| DeGLA2026.04 | 34.26 | |
| D_lc + InfoNCEData=D_lc, Loss=InfoNCE2026.04 | 34.15 | |
| D_wo + InfoNCEData=D_wo, Loss=InfoNCE2026.04 | 34.15 | |
| NegCLIP2026.04 | 34.11 | |
| TSVLC+2026.04 | 33.83 | |
| TSVLC2026.04 | 33.77 | |
| Slipform2026.04 | 33.64 | |
| Static Margin2026.04 | 33.45 | |
| D_lc + CementData=D_lc, Loss=Cement2026.04 | 33.42 | |
| Inverse Margin2026.04 | 33.33 | |
| D_hc + InfoNCEData=D_hc, Loss=InfoNCE2026.04 | 33.18 | |
| D_wo + CementData=D_wo, Loss=Cement2026.04 | 33.15 | |
| CLIP2026.04 | 33.08 | |
| TripletCLIP2026.04 | 32.11 |