Image Classification on ImageNet-Sketch (Accuracy and ECE)
68.6AccuracyMVT + FT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MVT + FTArch=ViT-g2025.12 | 68.6 | — | |
| Vanilla FTArch=ViT-g2025.12 | 68.1 | — | |
| MVTArch=ViT-g2025.12 | 67.9 | — | |
| EVAArch=ViT-g2025.12 | 67.7 | — | |
| TRACERBackbone=CLIP ViT-L/142026.05 | 63.71 | 10.52 | |
| CaRotBackbone=CLIP ViT-L/142026.05 | 62.66 | 9.14 | |
| MVT + FTMLLM=MMICL, Backbone=ViT-L2025.12 | 61.4 | — | |
| MVTMLLM=MMICL, Backbone=CLIP ViT-L, Fine-tuned=true2025.12 | 61.4 | — | |
| MVT + FTArch=ViT-L2025.12 | 61.4 | — | |
| FLYPBackbone=CLIP ViT-L/142026.05 | 60.2 | 21.64 | |
| MVT + FTMLLM=Otter, Backbone=ViT-L2025.12 | 60 | — | |
| MVTMLLM=Otter, Backbone=CLIP ViT-L, Fine-tuned=true2025.12 | 60 | — | |
| Vanilla FTArch=ViT-L2025.12 | 60 | — | |
| ZSBackbone=CLIP ViT-L/14, Protocol=Zero-Shot2026.05 | 59.61 | 10.37 | |
| MVTMLLM=Otter, Backbone=ViT-L2025.12 | 59.5 | — | |
| MVTMLLM=Otter, Backbone=CLIP ViT-L2025.12 | 59.5 | — | |
| CLIPMLLM=None, Backbone=ViT-L2025.12 | 59.4 | — | |
| CLIPMLLM=None, Backbone=CLIP ViT-L2025.12 | 59.4 | — | |
| CLIPArch=ViT-L2025.12 | 59.4 | — | |
| MVTMLLM=MMICL, Backbone=ViT-L2025.12 | 59 | — | |
| MVTMLLM=MMICL, Backbone=CLIP ViT-L2025.12 | 59 | — | |
| MVTArch=ViT-L2025.12 | 59 | — | |
| VQAArch=ViT-g2025.12 | 58.2 | — | |
| VQAArch=ViT-L2025.12 | 56.9 | — | |
| LP-FTBackbone=CLIP ViT-L/14, Protocol=Linear Probing then Fine-Tuning2026.05 | 56.84 | 26.13 | |
| FTBackbone=CLIP ViT-L/14, Protocol=Fine-Tuning2026.05 | 54.44 | 32.34 | |
| MERGETUNE + Weight ens.Backbone=CLIP ViT-B/16, Protocol=E2E-FT2026.01 | 53.1 | — | |
| VRFBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.01 | 52.93 | — | |
| MERGETUNEBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.01 | 52.76 | — | |
| TIESBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.01 | 51.67 | — | |
| Weight ens.Backbone=CLIP ViT-B/16, Protocol=E2E-FT2026.01 | 51.62 | — | |
| MERGETUNE + Weight ens.Backbone=CLIP ViT-B/16, Protocol=Linear Probing2026.01 | 50.14 | — | |
| PromptSRC2025.03 | 49.55 | — | |
| PromptSRC2025.05 | 49.55 | — | |
| MERGETUNEBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.01 | 49.47 | — | |
| MMRL++2025.05 | 49.3 | — | |
| MMRL2025.03 | 49.17 | — | |
| MMRL2025.05 | 49.17 | — | |
| MaPLe2025.03 | 49.15 | — | |
| MaPLe2025.05 | 49.15 | — | |
| MMA2025.03 | 49.13 | — | |
| MMA2025.05 | 49.13 | — | |
| CoCoOp2025.03 | 48.75 | — | |
| Co-CoOp2025.05 | 48.75 | — | |
| VRFBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.01 | 48.67 | — | |
| Weight ens.Backbone=CLIP ViT-B/16, Protocol=Linear Probing2026.01 | 48.41 | — | |
| Zero-shot (CLIP)Backbone=CLIP ViT-B/16, Protocol=Zero-shot2026.01 | 48.27 | — | |
| TIESBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.01 | 48.25 | — | |
| DAREBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.01 | 48.14 | — | |
| CoOp2025.03 | 47.99 | — | |
| CoOp2025.05 | 47.99 | — | |
| C-TPTBackbone=ViT-B/162025.03 | 47.9 | 7.35 | |
| MVTArch=ViT-B/32, Supervisor=MMICL [280]2025.12 | 47.8 | — | |
| TPTBackbone=ViT-B/162025.03 | 47.5 | 16.1 | |
| O-TPTBackbone=ViT-B/162025.03 | 47.12 | 6.87 | |
| Linear ProbingBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.01 | 46.99 | — | |
| CLIP2025.03 | 46.15 | — | |
| CLIP2025.05 | 46.15 | — | |
| CLIPBackbone=ViT-B/162025.03 | 46.1 | 4.95 | |
| DAREBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.01 | 45.79 | — | |
| E2E-FTBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.01 | 45.12 | — | |
| MVTArch=ViT-B/32, Fine-tuning=true, Supervisor=MMICL [280]2025.12 | 43.1 | — | |
| CLIPArch=ViT-B/322025.12 | 41.9 | — | |
| CLIPArch=ViT-B2025.12 | 41.9 | — | |
| MVTArch=ResNet-50, Supervisor=MMICL [280]2025.12 | 41.6 | — | |
| CLIPArch=RN1012025.12 | 40.9 | — | |
| FPLBackbone=ResNet-50, Evaluation protocol=Domain Generalization2026.01 | 36.78 | — | |
| MVTArch=ResNet-50, Fine-tuning=true, Supervisor=MMICL [280]2025.12 | 36.5 | — | |
| ZSBackbone=CLIP ResNet502026.05 | 35.45 | 7.4 | |
| Zero-Shot CLIPBackbone=ResNet-50, Evaluation protocol=Domain Generalization2026.01 | 35.44 | — | |
| CLIPArch=ResNet-502025.12 | 35.4 | — | |
| CLIPArch=RN502025.12 | 35.4 | — | |
| TPTBackbone=RN502025.03 | 35.1 | 13.7 | |
| C-TPTBackbone=RN502025.03 | 35.1 | 9.7 | |
| TPTBackbone=ResNet-50, Evaluation protocol=Domain Generalization2026.01 | 35.09 | — | |
| CoOpBackbone=ResNet-50, Evaluation protocol=Domain Generalization2026.01 | 34.67 | — | |
| CoCoOpBackbone=ResNet-50, Evaluation protocol=Domain Generalization2026.01 | 34.48 | — | |
| ProGradBackbone=ResNet-50, Evaluation protocol=Domain Generalization2026.01 | 34.4 | — | |
| TRACERBackbone=CLIP ResNet502026.05 | 34.34 | 17.41 | |
| FLYPBackbone=CLIP ResNet502026.05 | 34.24 | 20.21 | |
| CaRotBackbone=CLIP ResNet502026.05 | 34.05 | 34.35 | |
| O-TPTBackbone=RN502025.03 | 33.98 | 5.85 | |
| FTBackbone=CLIP ResNet502026.05 | 33.9 | 28.24 | |
| LP-FTBackbone=CLIP ResNet502026.05 | 33.33 | 35.2 | |
| ZERO-SHOT CLIPMode=Zero-shot2025.12 | 33.32 | — | |
| CLIPBackbone=RN502025.03 | 33.3 | 3.15 | |
| PLOTBackbone=ResNet-50, Evaluation protocol=Domain Generalization2026.01 | 33 | — | |
| RETROPROMPT2025.12 | 32.89 | — | |
| COOP2025.12 | 32.74 | — |