Fine-grained Image Classification on 8 Fine-grained Dataset Suite Average
93.4RobustnessAGC
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AGCBackbone=ViT-L/14, Epsilon=4/2552026.05 | 93.4 | — | 68.8 | — | |
| AGCBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 91.4 | 60.7 | — | — | |
| AGCBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05 | 91.4 | 60.7 | — | — | |
| AGCBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 82.9 | 56.7 | — | — | |
| AGCBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.05 | 70.6 | 53.5 | — | — | |
| EnsembleBackbone=ViT-L/14, Epsilon=4/2552026.05 | 52.9 | — | 67.7 | — | |
| TTPBackbone=ViT-L/14, Epsilon=4/2552026.05 | 51.6 | — | 68.9 | — | |
| R-TPTBackbone=ViT-L/14, Epsilon=4/2552026.05 | 49.6 | — | 68.4 | — | |
| APT+TeCoABackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.05 | 46.4 | 46.5 | — | — | |
| TTPBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.05 | 46.2 | 53 | — | — | |
| R-TPTBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.05 | 45.2 | 53.7 | — | — | |
| TTPBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 43 | 61.2 | — | — | |
| TTPBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05 | 43 | 61.2 | — | — | |
| EnsembleBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 42.7 | 59.3 | — | — | |
| EnsembleBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05 | 42.7 | 59.3 | — | — | |
| EnsembleBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 40 | 55.9 | — | — | |
| MTABackbone=ViT-L/14, Epsilon=4/2552026.05 | 39.9 | — | 69.1 | — | |
| R-TPTBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 39.9 | 61.1 | — | — | |
| R-TPTBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05 | 39.9 | 61.1 | — | — | |
| TTPBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 39.7 | 57.1 | — | — | |
| TeCoABackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.05 | 37 | 37.1 | — | — | |
| R-TPTBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 35.3 | 57.1 | — | — | |
| MTABackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 35 | 58.5 | — | — | |
| MTABackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 27.4 | 62.4 | — | — | |
| MTABackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05 | 27.4 | 62.4 | — | — | |
| T-VSSBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 22 | — | 30.9 | — | |
| R-TPTBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 21.2 | — | 30.9 | — | |
| EnsembleBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 19.5 | — | 28.9 | — | |
| MTABackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 17.8 | — | 32.9 | — | |
| TPTBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 16.5 | — | 32.7 | — | |
| C-TPTBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 13.9 | — | 33.5 | — | |
| CLIP-TeCoABackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 12.3 | — | 33.4 | — | |
| TTCBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 6.9 | 56.7 | — | — | |
| TTCBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 4.5 | 58.9 | — | — | |
| TTCBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05 | 4.5 | 58.9 | — | — | |
| TTCBackbone=ViT-L/14, Epsilon=4/2552026.05 | 4.3 | — | 68.2 | — | |
| APTBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.05 | 2.6 | 11.4 | — | — | |
| CLIPBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.05 | 0.4 | 53.2 | — | — | |
| CLIPBackbone=ViT-L/14, Epsilon=4/2552026.05 | 0 | — | 69.1 | — | |
| CLIPBackbone=ViT-B/32, Epsilon (ε)=4/2552026.05 | 0 | 57.4 | — | — | |
| CLIPBackbone=ViT-B/16, Epsilon (ε)=4/2552026.05 | 0 | 61.4 | — | — | |
| CLIPBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05 | 0 | 61.4 | — | — | |
| A-TPTStage=Test time, Backbone=ViT-B/322026.05 | — | — | 58.6 | 40.3 | |
| APT+TeCoAStage=Training time, Backbone=ViT-B/322026.05 | — | — | 40.2 | 38.9 | |
| FAREStage=Training time, Backbone=ViT-B/322026.05 | — | — | 42.2 | 40.3 | |
| PAFTStage=Training time, Backbone=ViT-B/322026.05 | — | — | 54.9 | 27.5 | |
| TeCoAStage=Training time, Backbone=ViT-B/322026.05 | — | — | 33.4 | 32.2 |