Image Classification on ImageNet-A (OOD Evaluation)
76.84AccuracyCoOpOp
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| CoOpOp2026.05 | 76.84 | — | — | — | — | — | |
| RPO2026.05 | 76.83 | — | — | — | — | — | |
| MVT + FTArch=ViT-g2025.12 | 75.5 | — | — | — | — | — | |
| CLIP-Adapter2026.05 | 75.39 | — | — | — | — | — | |
| MVTMLLM=MMICL, Backbone=CLIP ViT-L, Fine-tuned=true2025.12 | 75.1 | — | — | — | — | — | |
| MVT + FTArch=ViT-L2025.12 | 75.1 | — | — | — | — | — | |
| MVTMLLM=Otter, Backbone=CLIP ViT-L, Fine-tuned=true2025.12 | 73.5 | — | — | — | — | — | |
| MVTArch=ViT-g2025.12 | 73.2 | — | — | — | — | — | |
| Vanilla FTArch=ViT-g2025.12 | 72.7 | — | — | — | — | — | |
| EVAArch=ViT-g2025.12 | 71.9 | — | — | — | — | — | |
| MVTMLLM=MMICL, Backbone=CLIP ViT-L2025.12 | 71.2 | — | — | — | — | — | |
| MVTArch=ViT-L2025.12 | 71.2 | — | — | — | — | — | |
| Vanilla FTArch=ViT-L2025.12 | 70.8 | — | — | — | — | — | |
| CLIPMLLM=None, Backbone=CLIP ViT-L2025.12 | 69.3 | — | — | — | — | — | |
| CLIPArch=ViT-L2025.12 | 69.3 | — | — | — | — | — | |
| VQAArch=ViT-L2025.12 | 66 | — | — | — | — | — | |
| VQAArch=ViT-g2025.12 | 64.6 | — | — | — | — | — | |
| MVTMLLM=Otter, Backbone=CLIP ViT-L2025.12 | 64.1 | — | — | — | — | — | |
| Fair Context Learning2026.02 | 61.38 | 62.91 | — | — | — | — | |
| ZERO2026.02 | 59.77 | 62.48 | — | — | — | — | |
| TTL2026.02 | 59 | 62.47 | — | — | — | — | |
| TPS2026.02 | 58.19 | 61.77 | — | — | — | — | |
| R-TPTAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/162026.06 | 57.9 | — | 20.7 | — | — | — | |
| R-TPT2026.02 | 57.72 | 61.6 | — | — | — | — | |
| RLCF2026.02 | 57.39 | 61.37 | — | — | — | — | |
| MTA2026.02 | 57.03 | 61.53 | — | — | — | — | |
| TPT2026.02 | 54.75 | 60.8 | — | — | — | — | |
| PCSRadaptation_mode=test-time adaptation2025.12 | 52.1 | — | — | — | — | — | |
| DBDAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/162026.06 | 52.1 | — | 98.9 | — | — | — | |
| C-TPT2026.02 | 51.35 | 59.33 | — | — | — | — | |
| CLIPAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/162026.06 | 50.6 | — | 0 | — | — | — | |
| CLIP2026.02 | 47.8 | 57.2 | — | — | — | — | |
| TTCAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/162026.06 | 46.6 | — | 9.9 | — | — | — | |
| SARadaptation_mode=test-time adaptation2025.12 | 45.3 | — | — | — | — | — | |
| SFT-EMModel=Qwen2.5-VL-7B2026.02 | 44.57 | 62.1 | — | — | — | — | |
| TENTadaptation_mode=test-time adaptation2025.12 | 44.5 | — | — | — | — | — | |
| GRPOModel=Qwen2.5-VL-7B2026.02 | 43.65 | 59.48 | — | — | — | — | |
| SFT-MModel=Qwen2.5-VL-7B2026.02 | 42.49 | 60.41 | — | — | — | — | |
| SFTModel=Qwen2.5-VL-7B2026.02 | 39.14 | 57.62 | — | — | — | — | |
| MTACategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack scenario=PGD-100 (eps=4.0)2026.06 | 37.9 | — | 5.5 | — | — | — | |
| SFT-MModel=Qwen2.5-VL-3B2026.02 | 37.61 | 59.02 | — | — | — | — | |
| SFT-EMModel=Qwen2.5-VL-3B2026.02 | 37.25 | 59.44 | — | — | — | — | |
| R-TPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack scenario=PGD-100 (eps=4.0)2026.06 | 37.1 | — | 11.7 | — | — | — | |
| R-TPTAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/322026.06 | 36.6 | — | 11 | — | — | — | |
| NAMEx-Full-MomParams=50M, Momentum=true, Zero-shot=true2025.10 | 35.27 | — | — | — | — | — | |
| GRPOModel=Qwen2.5-VL-3B2026.02 | 35.16 | 58.31 | — | — | — | — | |
| NAMEx-MomParams=50M, Momentum=true, Zero-shot=true2025.10 | 35.05 | — | — | — | — | — | |
| SFTModel=Qwen2.5-VL-3B2026.02 | 34.63 | 56.5 | — | — | — | — | |
| EP-CAMEx-MomParams=50M, Momentum=true, Zero-shot=true2025.10 | 33.22 | — | — | — | — | — | |
| DBDAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/322026.06 | 32.9 | — | 95.3 | — | — | — | |
| CLIPArch=ViT-B2025.12 | 32.2 | — | — | — | — | — | |
| DOCOSetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 31.7 | — | — | — | — | — | |
| E-COMESetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 31.2 | — | — | — | — | — | |
| Sourceadaptation_mode=test-time adaptation2025.12 | 31.1 | — | — | — | — | — | |
| CLIPAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/322026.06 | 31.1 | — | 0 | — | — | — | |
| CLIPCategory=Baseline, Backbone=CLIP-ViT-B/32, Attack scenario=PGD-100 (eps=4.0)2026.06 | 30.4 | — | 0 | — | — | — | |
| CLIPArch=RN1012025.12 | 30.2 | — | — | — | — | — | |
| TTCCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack scenario=PGD-100 (eps=4.0)2026.06 | 30 | — | 0.6 | — | — | — | |
| EATASetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 29.9 | — | — | — | — | — | |
| TentSetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 29.8 | — | — | — | — | — | |
| OSTTASetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 29.7 | — | — | — | — | — | |
| ViDASetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 29.5 | — | — | — | — | — | |
| UniEntSetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 29.3 | — | — | — | — | — | |
| DPCoreSetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 29.3 | — | — | — | — | — | |
| MACCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack scenario=PGD-100 (eps=4.0)2026.06 | 29.1 | — | 14.8 | — | — | — | |
| STAMPSetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 28.8 | — | — | — | — | — | |
| R-TPTBackbone=CLIP-ResNet502025.04 | 28.4 | — | 14.4 | — | — | — | |
| CoTTASetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 28.2 | — | — | — | — | — | |
| SourceSetting=Closed-set, Scenario=Continual Test-Time Adaptation2026.04 | 28.1 | — | — | — | — | — | |
| R-TPTAttack=PGD-10, Epsilon=1/255, Backbone=CLIP-ResNet502026.06 | 28 | — | 14.2 | — | — | — | |
| MTABackbone=CLIP-ResNet502025.04 | 27.5 | — | 5.6 | — | — | — | |
| TTCAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/322026.06 | 27.5 | — | 8.4 | — | — | — | |
| SS-TPT2026.06 | 26.9 | — | 14 | — | — | — | |
| TPTBackbone=CLIP-ResNet502025.04 | 26.5 | — | 0 | — | — | — | |
| R-TPT2026.06 | 26 | — | 10.9 | — | — | — | |
| TPT2026.06 | 25.9 | — | 0.1 | — | — | — | |
| NAMEx-FullParams=50M, Momentum=false, Zero-shot=true2025.10 | 25.74 | — | — | — | — | — | |
| CAMExParams=50M, Zero-shot=true2025.10 | 25.45 | — | — | — | — | — | |
| NAMExParams=50M, Momentum=false, Zero-shot=true2025.10 | 25.32 | — | — | — | — | — | |
| EP-CAMExParams=50M, Momentum=false, Zero-shot=true2025.10 | 24.22 | — | — | — | — | — | |
| SMEARParams=50M, Zero-shot=true2025.10 | 24.14 | — | — | — | — | — | |
| C-TPTBackbone=CLIP-ResNet502025.04 | 24.1 | — | 0 | — | — | — | |
| CLIPArch=RN502025.12 | 23.9 | — | — | — | — | — | |
| CLIPAttack=PGD-10, Epsilon=1/255, Backbone=CLIP-ResNet502026.06 | 23.8 | — | 0 | — | — | — | |
| SMoEParams=50M, Zero-shot=true2025.10 | 23.72 | — | — | — | — | — | |
| DBDAttack=PGD-10, Epsilon=1/255, Backbone=CLIP-ResNet502026.06 | 23 | — | 89 | — | — | — | |
| EnsembleBackbone=CLIP-ResNet502025.04 | 22.6 | — | 10.1 | — | — | — | |
| Ensemble2026.06 | 22.6 | — | 6.5 | — | — | — | |
| CLIPBackbone=CLIP-ResNet502025.04 | 21.8 | — | 0 | — | — | — | |
| CLIP2026.06 | 21.8 | — | 0 | — | — | — | |
| PiT-S2021.03 | 21.7 | — | — | — | — | — | |
| TTC2026.06 | 21.3 | — | 1.1 | — | — | — | |
| DOC2026.06 | 21.1 | — | 1.1 | — | — | — | |
| ViT-S2021.03 | 19.1 | — | — | — | — | — | |
| ResNet50variant=well-optimized (†)2021.03 | 5.4 | — | — | — | — | — | |
| TeCoAAttack=PGD-10, Epsilon=1/255, Backbone=CLIP-ResNet502026.06 | 4.9 | — | 1.2 | — | — | — | |
| TeCoAAttack=PGD-100, Epsilon=4/255, Backbone=CLIP-ViT-B/322026.06 | 4.2 | — | 0.3 | — | — | — | |
| ResNet502021.03 | 0 | — | — | — | — | — | |
| A-VARCModel size=L/16, Model type=Autoregressive2026.03 | — | — | — | — | — | 8.2 | |
| A-VARCModel size=XL/16, Model type=Autoregressive2026.03 | — | — | — | — | — | 9 |