Fine-grained Image Classification on Describable Textures Dataset (DTD)
81.54Top-1 Accuracy (Clean)C-RADIOv3-B
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| C-RADIOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 81.54 | — | — | — | — | |
| DFN2B-CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 80.74 | — | — | — | — | |
| DINOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 80.32 | — | — | — | — | |
| DINOv2-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 79.47 | — | — | — | — | |
| OpenCLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 79.41 | — | — | — | — | |
| DINOv2-Reg-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 79.2 | — | — | — | — | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=642026.05 | 77.01 | — | — | — | — | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=82026.05 | 75.8 | — | — | — | — | |
| SigLIP2-B/16-256Evaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 75.32 | — | — | — | — | |
| CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 74.15 | — | — | — | — | |
| Clean (Reference)Backbone=CLIP-ViT-B/16, Labeling Strategy=none (clean accuracy)2026.06 | 53.2 | — | — | — | — | |
| WISE-FTBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 42.07 | — | — | 0 | — | |
| GRACEBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 41.6 | — | — | 12.6 | — | |
| CLIPBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.03 | 41.43 | — | — | 0 | — | |
| TPGMBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 40.85 | — | — | 0 | — | |
| SPDBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 40.53 | — | — | 0 | — | |
| Vanilla FTBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 40.26 | — | — | 0 | — | |
| FLYPBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 38.56 | — | — | 0 | — | |
| PMG-AFTBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 30.2 | — | — | 12.9 | — | |
| LAATBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 29.9 | — | — | 11.8 | — | |
| FAREBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 27.87 | — | — | 12.2 | — | |
| TeCoABackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 24.04 | — | — | 10.7 | — | |
| APTBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | — | 9 | — | — | 0.1 | |
| APT+TeCoABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | — | 35.2 | — | — | 33.7 | |
| C-TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | — | 45.9 | 0 | — | — | |
| CLIPBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | — | 44.4 | 0 | — | — | |
| CLIPBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06 | — | — | — | — | 4.2 | |
| CLIPBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Baseline2026.06 | — | 43 | — | — | 0 | |
| DBDBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06 | — | — | — | — | 53.1 | |
| EnsembleBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | — | 43.2 | 25.1 | — | — | |
| EnsembleBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | — | 39.8 | — | — | 28.6 | |
| FAREBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | — | 28.3 | — | — | 27.3 | |
| MTABackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | — | 46.5 | 16.2 | — | — | |
| MTABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | — | 43.8 | — | — | 28.8 | |
| R-TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | — | 46.4 | 32.8 | — | — | |
| R-TPTBackbone=CLIP-ViT-B/16, Attack Type=PGD-100, Perturbation Budget (epsilon)=4/255, Labeling Strategy=pseudo-labels2026.06 | — | — | — | — | 37.1 | |
| R-TPTBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | — | 42.1 | — | — | 29.1 | |
| T-VSSBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | — | 43.3 | — | — | 33.1 | |
| TeCoABackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Training-time Defense2026.06 | — | 24.5 | — | — | 24 | |
| TPTBackbone=CLIP-ViT-B/16, Epsilon (ε)=4.02025.04 | — | 46.9 | 0 | — | — | |
| TTCBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | — | 37.3 | — | — | 4.7 | |
| TTPBackbone=CLIP-ViT-B/32, Epsilon (ϵ)=4/255, Defense Category=Test-time Defense2026.06 | — | 42.8 | — | — | 32.2 |