Fine-grained Image Classification on Oxford Flowers 102
98.9AccuracyDATL + WS-DAN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DATL + WS-DAN2020.10 | 98.9 | — | — | |
| Sequential-to-global SSL frameworkEvaluation Protocol=Linear Probing, Operating Mode=Static ViT2026.04 | 98.3 | — | — | |
| Sequential-to-global SSL frameworkEvaluation Protocol=Linear Probing, Operating Mode=Learned Policy, Iterative Steps=82026.04 | 97.7 | — | — | |
| Sequential-to-global SSL frameworkEvaluation Protocol=Linear Probing, Operating Mode=Random Gazes, Iterative Steps=82026.04 | 97.6 | — | — | |
| WS-DAN2020.10 | 97.1 | — | — | |
| DINOEvaluation Protocol=Linear Probing, Operating Mode=Static ViT2026.04 | 96.4 | — | — | |
| ToolFGshot=42026.06 | 95.8 | — | — | |
| PromptSRCshot=42026.06 | 92.4 | — | — | |
| DiVE-kshot=42026.06 | 88.7 | — | — | |
| Fine-R1shot=42026.06 | 86 | — | — | |
| ViRFTshot=42026.06 | 84.8 | — | — | |
| MaPLeshot=42026.06 | 84.6 | — | — | |
| Qwen2.5-VL-7Bshot=42026.06 | 78.4 | — | — | |
| Clean (Reference)Backbone=CLIP-ViT-B/16, Labeling Strategy=none (clean accuracy)2026.06 | 73 | — | — | |
| C-TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | 69.6 | 0 | — | |
| TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | 69.1 | 0 | — | |
| TTCBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 69 | — | 7.4 | |
| R-TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | 68.7 | 44.6 | — | |
| R-TPTBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 68.7 | — | 44.6 | |
| AGCBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 67.8 | — | 89.6 | |
| CLIPBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | 67.4 | 0 | — | |
| MTABackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | 67.4 | 27.9 | — | |
| CLIPBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 67.4 | — | 0 | |
| MTABackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 67.4 | — | 27.9 | |
| TTPBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 67.3 | — | 47.2 | |
| EnsembleBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 67.2 | — | 46.9 | |
| EnsembleBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | 65.9 | 36.3 | — | |
| MTABackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 64.4 | — | 36.5 | |
| MTABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | 64.4 | — | 36.5 | |
| TTCBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 64.3 | — | 3.2 | |
| TTCBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | 64.3 | — | 3.2 | |
| CLIPBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 64 | — | 0 | |
| CLIPBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Baseline2026.06 | 64 | — | 0 | |
| TTPBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 63.6 | — | 42 | |
| TTPBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | 63.6 | — | 42 | |
| AGCBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 63.1 | — | 71.5 | |
| R-TPTBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 62.6 | — | 37.6 | |
| R-TPTBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | 62.6 | — | 37.6 | |
| T-VSSBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | 62.4 | — | 45.2 | |
| EnsembleBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 61.7 | — | 43.2 | |
| EnsembleBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | 58.1 | — | 36.1 | |
| APT+TeCoABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | 42.5 | — | 40.4 | |
| FAREBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | 37 | — | 34 | |
| TeCoABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | 30.8 | — | 28.9 | |
| APTBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | 0.9 | — | 0.2 | |
| CLIPBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06 | — | — | 2.7 | |
| DBDBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06 | — | — | 72.3 | |
| R-TPTBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06 | — | — | 49.8 |