Image Classification on ImageNet OOD
71.51ImageNet AccCoOp
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CoOpBackbone=ViT-B/16, T.D=false, Avail.=false, T.F.=true2025.07 | 71.51 | 49.71 | 64.2 | 75.21 | 47.99 | 61.72 | 59.28 | — | — | |
| CoCoOpBackbone=ViT-B/16, T.D=false, Avail.=false, T.F.=true2025.07 | 71.02 | 50.63 | 64.07 | 76.18 | 48.75 | 62.13 | 59.91 | — | — | |
| Tip-AdapterBackbone=ViT-B/16, T.D=false, Avail.=false, T.F.=true2025.07 | 70.75 | 51.04 | 63.41 | 77.76 | 48.88 | 62.37 | 60.27 | — | — | |
| DiffTPTBackbone=ViT-B/16, T.D=true, Avail.=false, T.F.=false2025.07 | 70.3 | 55.68 | 65.1 | 75 | 46.8 | 62.28 | 60.52 | — | — | |
| DiffTPTVisual Backbone=ViT-B/16, Bp-free=false2025.03 | 70.3 | 55.68 | 65.1 | 75 | 46.8 | 62.28 | 60.52 | — | — | |
| BCAVisual Backbone=ViT-B/16, Bp-free=true2025.03 | 70.22 | 61.14 | 64.9 | 80.72 | 50.87 | 65.37 | 64.16 | — | — | |
| FreeTTABackbone=ViT-B/16, T.D=true, Avail.=true, T.F.=false2025.07 | 70.21 | 61.41 | 64.92 | 80.49 | 50.88 | 65.58 | 64.42 | — | — | |
| MTAVisual Backbone=ViT-B/16, Bp-free=true2025.03 | 70.08 | 58.06 | 64.24 | 78.33 | 49.61 | 64.06 | 62.56 | — | — | |
| TDABackbone=ViT-B/16, T.D=true, Avail.=false, T.F.=false2025.07 | 69.51 | 60.11 | 64.67 | 80.24 | 50.54 | 65.01 | 63.89 | — | — | |
| TDAVisual Backbone=ViT-B/16, Bp-free=true2025.03 | 69.51 | 60.11 | 64.67 | 80.24 | 50.54 | 65.01 | 63.89 | — | — | |
| C-TPTVisual Backbone=ViT-B/16, Bp-free=false2025.03 | 69.3 | 52.9 | 63.4 | 78 | 48.5 | 62.4 | 60.7 | — | — | |
| MTABackbone=ViT-B/16, T.D=true, Avail.=true, T.F.=false2025.07 | 69.29 | 57.41 | 63.61 | 76.92 | 48.58 | 63.16 | 61.63 | — | — | |
| CLIP with PAVisual Backbone=ViT-B/16, Bp-free=true2025.03 | 69.16 | 59.39 | 64.21 | 79.68 | 49.88 | 64.46 | 63.29 | — | — | |
| ZEROBackbone=ViT-B/16, T.D=true, Avail.=true, T.F.=false2025.07 | 69.06 | 61.35 | 64.13 | 77.28 | 48.29 | 64.02 | 62.76 | — | — | |
| CLIP with LAVisual Backbone=ViT-B/16, Bp-free=true2025.03 | 68.99 | 58.62 | 63.64 | 79.71 | 49.58 | 64.11 | 62.89 | — | — | |
| TPTBackbone=ViT-B/16, T.D=true, Avail.=false, T.F.=false2025.07 | 68.98 | 54.77 | 63.45 | 77.06 | 47.94 | 62.44 | 60.81 | — | — | |
| TPTVisual Backbone=ViT-B/16, Bp-free=false2025.03 | 68.98 | 54.77 | 63.45 | 77.06 | 47.94 | 62.44 | 60.81 | — | — | |
| CLIP-ViT-B/16KeepRate=0.7, GFLOPs=17.62026.03 | 68.37 | 50.2 | 61.86 | 77.55 | — | 61.24 | — | 48.24 | — | |
| CLIPBackbone=ViT-B/162025.07 | 68.34 | 49.89 | 61.88 | 77.65 | 48.24 | 61.2 | 59.42 | — | — | |
| CLIPVisual Backbone=ViT-B/16, Bp-free=true2025.03 | 68.34 | 49.89 | 61.88 | 77.65 | 48.24 | 61.2 | 59.42 | — | — | |
| TCA (Baseline)Layers=3, 6, 9, KeepRate=0.7, GFLOPs=11.92026.03 | 65.46 | 47.42 | 59.14 | 66.04 | — | 56.39 | — | 43.91 | — | |
| Col-LnLayers=3, 6, 9, KeepRate=0.7, GFLOPs=11.92026.03 | 65.31 | 49.27 | 59.13 | 72.1 | — | 58.39 | — | 45.64 | — | |
| CoOpBackbone=ResNet-50, T.D=false, Avail.=false, T.F.=true2025.07 | 63.33 | 23.06 | 55.4 | 56.6 | 34.67 | 46.61 | 42.43 | — | — | |
| FLYPEnsembling=true2022.12 | 63.2 | — | — | — | — | — | — | — | — | |
| TOGA2026.03 | 63.1 | — | — | — | — | — | — | — | — | |
| CoCoOpBackbone=ResNet-50, T.D=false, Avail.=false, T.F.=true2025.07 | 62.81 | 23.32 | 55.72 | 57.74 | 34.48 | 46.81 | 42.82 | — | — | |
| EVILBackbone=CLIP ViT-B/16, Protocol=Sparse Training2025.12 | 62.5 | — | — | — | — | — | — | — | — | |
| MaPLe2026.03 | 62.4 | — | — | — | — | — | — | — | — | |
| KgCoOp2026.03 | 62.3 | — | — | — | — | — | — | — | — | |
| TCA (Baseline)Layers=0, 3, 6, KeepRate=0.7, GFLOPs=9.22026.03 | 62.26 | 38.97 | 55.71 | 60.43 | — | 51.69 | — | 41.09 | — | |
| Col-LnLayers=0, 3, 6, KeepRate=0.7, GFLOPs=9.22026.03 | 62.25 | 40.47 | 55.51 | 64.33 | — | 52.73 | — | 41.09 | — | |
| CoCoOp2026.03 | 62.1 | — | — | — | — | — | — | — | — | |
| Tip-AdapterBackbone=ResNet-50, T.D=false, Avail.=false, T.F.=true2025.07 | 62.03 | 23.13 | 53.97 | 60.35 | 35.74 | 47.04 | 43.3 | — | — | |
| BCAVisual Backbone=ResNet-50, Bp-free=true2025.03 | 61.81 | 30.35 | 56.58 | 62.89 | 38.08 | 49.94 | 46.98 | — | — | |
| LP-FTEnsembling=true2022.12 | 61.8 | — | — | — | — | — | — | — | — | |
| VRFBackbone=CLIP ViT-B/16, Evaluation Protocol=Selective prediction using OOD detector2024.11 | 61.8 | — | — | — | — | — | — | — | — | |
| ProGrad2026.03 | 61.7 | — | — | — | — | — | — | — | — | |
| FreeTTABackbone=ResNet-50, T.D=true, Avail.=true, T.F.=false2025.07 | 61.51 | 30.67 | 55.89 | 63.02 | 37.94 | 49.81 | 46.88 | — | — | |
| TransCLIP2026.03 | 61.4 | — | — | — | — | — | — | — | — | |
| TDAVisual Backbone=ResNet-50, Bp-free=true2025.03 | 61.35 | 30.29 | 55.54 | 62.58 | 38.12 | 49.58 | 46.63 | — | — | |
| FTEnsembling=true2022.12 | 61.3 | — | — | — | — | — | — | — | — | |
| FinetuningBackbone=CLIP ViT-B/16, Protocol=Finetuning2025.12 | 61.3 | — | — | — | — | — | — | — | — | |
| C-TPTVisual Backbone=ResNet-50, Bp-free=false2025.03 | 61.2 | 25.6 | 54.8 | 59.7 | 35.7 | 47.4 | 44 | — | — | |
| CLIP with PAVisual Backbone=ResNet-50, Bp-free=true2025.03 | 61.15 | 28.66 | 55.41 | 61.97 | 37.41 | 48.92 | 45.86 | — | — | |
| CoOp2026.03 | 61.1 | — | — | — | — | — | — | — | — | |
| CLIP with LAVisual Backbone=ResNet-50, Bp-free=true2025.03 | 60.95 | 28.53 | 55.59 | 61.88 | 37.21 | 48.83 | 45.8 | — | — | |
| DiffTPTBackbone=ResNet-50, T.D=true, Avail.=false, T.F.=false2025.07 | 60.8 | 31.06 | 55.8 | 58.8 | 37.1 | 48.71 | 45.69 | — | — | |
| DiffTPTVisual Backbone=ResNet-50, Bp-free=false2025.03 | 60.8 | 31.06 | 55.8 | 58.8 | 37.1 | 48.71 | 45.69 | — | — | |
| TPTBackbone=ResNet-50, T.D=true, Avail.=false, T.F.=false2025.07 | 60.74 | 26.67 | 54.7 | 59.11 | 35.09 | 47.26 | 43.89 | — | — | |
| TPTVisual Backbone=ResNet-50, Bp-free=false2025.03 | 60.74 | 26.67 | 54.7 | 59.11 | 35.09 | 47.26 | 43.89 | — | — | |
| LP-FTEnsembling=false2022.12 | 60.5 | — | — | — | — | — | — | — | — | |
| FLYPEnsembling=false2022.12 | 60.2 | — | — | — | — | — | — | — | — | |
| CLIPBackbone=ResNet-502025.07 | 59.81 | 23.24 | 52.91 | 60.72 | 35.48 | 46.43 | 43.09 | — | — | |
| CLIPVisual Backbone=ResNet-50, Bp-free=true2025.03 | 59.81 | 23.24 | 52.91 | 60.72 | 35.48 | 46.43 | 43.09 | — | — | |
| CLIP2026.03 | 59.1 | — | — | — | — | — | — | — | — | |
| L2-SPEnsembling=true2022.12 | 58.9 | — | — | — | — | — | — | — | — | |
| ZeroshotEnsembling=false2022.12 | 58.7 | — | — | — | — | — | — | — | — | |
| ZeroshotEnsembling=true2022.12 | 58.7 | — | — | — | — | — | — | — | — | |
| ZeroshotBackbone=CLIP ViT-B/16, Protocol=Zero-shot2025.12 | 58.7 | — | — | — | — | — | — | — | — | |
| KNNBackbone=CLIP ViT-B/16, Evaluation Protocol=Selective prediction using OOD detector2024.11 | 58.4 | — | — | — | — | — | — | — | — | |
| RMDBackbone=CLIP ViT-B/16, Evaluation Protocol=Selective prediction using OOD detector2024.11 | 58.4 | — | — | — | — | — | — | — | — | |
| LPEnsembling=true2022.12 | 58.3 | — | — | — | — | — | — | — | — | |
| L2-SPEnsembling=false2022.12 | 57.9 | — | — | — | — | — | — | — | — | |
| MDBackbone=CLIP ViT-B/16, Evaluation Protocol=Selective prediction using OOD detector2024.11 | 57.7 | — | — | — | — | — | — | — | — | |
| EnergyBackbone=CLIP ViT-B/16, Evaluation Protocol=Selective prediction using OOD detector2024.11 | 57.6 | — | — | — | — | — | — | — | — | |
| MSPBackbone=CLIP ViT-B/16, Evaluation Protocol=Selective prediction using OOD detector2024.11 | 57.3 | — | — | — | — | — | — | — | — | |
| LPEnsembling=false2022.12 | 57.2 | — | — | — | — | — | — | — | — | |
| FTEnsembling=false2022.12 | 54.8 | — | — | — | — | — | — | — | — | |
| LP-FTBackbone=CLIP ViT-B/162023.03 | — | 49.1 | 71.6 | 72.9 | — | — | — | 48.4 | — | |
| Ours (multi-fill)Backbone=CLIP ViT-B/322023.03 | — | 33 | 67.7 | 68.1 | — | — | 52.6 | 46.6 | 47.5 | |
| Ours (multi-fill)Backbone=CLIP ViT-B/162023.03 | — | 52 | 73.4 | 76.4 | — | — | 62.3 | 52.7 | 56.8 | |
| Ours (single-fill)Backbone=CLIP ViT-B/322023.03 | — | 33.8 | 67.1 | 69.7 | — | — | 53.1 | 46.9 | 48 | |
| Ours (single-fill)Backbone=CLIP ViT-B/162023.03 | — | 53.5 | 73.4 | 78.1 | — | — | 63.3 | 53.4 | 57.9 | |
| PromptAlignBackbone=ViT-B/16, T.D=true, Avail.=false, T.F.=false2025.07 | — | 59.37 | 65.29 | 79.33 | 50.23 | — | 63.55 | — | — | |
| PromptAlignVisual Backbone=ViT-B/16, Bp-free=false2025.03 | — | 59.37 | 65.29 | 79.33 | 50.23 | — | 63.55 | — | — | |
| Uniform soupBackbone=CLIP ViT-B/322023.03 | — | 29.2 | 68.6 | 66.6 | — | — | 51.6 | 47.7 | 46.1 | |
| Vanilla fine-tuningBackbone=CLIP ViT-B/322023.03 | — | 20 | 64.7 | 57 | — | — | 44.2 | 39.8 | 39.5 | |
| Vanilla fine-tuningBackbone=CLIP ViT-B/162023.03 | — | 35.2 | 70.4 | 64 | — | — | 52.8 | 45.1 | 49.1 | |
| WiSE-FTBackbone=CLIP ViT-B/322023.03 | — | 31.9 | 66.6 | 70.2 | — | — | 52.4 | 47.1 | 46.3 | |
| WiSE-FTBackbone=CLIP ViT-B/162023.03 | — | 52.2 | 72.8 | 78.7 | — | — | 63 | 53.9 | 57.3 | |
| Zero-shotBackbone=CLIP ViT-B/322023.03 | — | 31.4 | 55.9 | 69.3 | — | — | 48.7 | 42.3 | 44.5 | |
| Zero-shotBackbone=CLIP ViT-B/162023.03 | — | 50.1 | 61.9 | 77.6 | — | — | 58.4 | 48.3 | 54 |