List your company
ResearchFormerly WizWand

Fine-grained Classification on Cars (Accuracy and Robustness)

78.4AccuracyMTA
2.63622.305541.97561.6445Dec 18, 2025Jan 15, 2026Feb 12, 2026Mar 12, 2026Apr 9, 2026May 7, 2026Jun 5, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MTABackbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.1278.436.6
MTABackbone=ViT-L/14, Epsilon=4/2552026.0578.436.6
MTACategory=Tuning-free, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0678.436.6
AGCBackbone=ViT-L/14, Epsilon=4/2552026.0577.395.4
R-TPTBackbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.1277.249.1
R-TPTBackbone=ViT-L/14, Epsilon=4/2552026.0577.249.1
R-TPTCategory=Tuning-based, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0677.249.1
CLIPBackbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.1276.80
TTPBackbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.1276.851.1
CLIPBackbone=ViT-L/14, Epsilon=4/2552026.0576.80
TTPBackbone=ViT-L/14, Epsilon=4/2552026.0576.851.1
CLIPCategory=Baseline, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0676.80
EnsembleBackbone=ViT-L/14, Epsilon=4/2552026.0576.654.4
EnsembleBackbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.1276.340.5
TTCCategory=Tuning-free, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0674.63.2
MACCategory=Tuning-free, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0674.669.8
TTCBackbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.1267.82.2
TTCBackbone=ViT-L/14, Epsilon=4/2552026.0567.82.2
MTABackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0567.718.5
AGCBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0567.393.2
R-TPTBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.056734.7
EnsembleBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.056641.8
CLIPBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0565.50
TTPBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0565.437.4
R-TPTBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0558.142.9
AGCBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0557.480.1
CLIPBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0555.70
TTPBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0555.641.5
TTCBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.05552.9
APT+TeCoABackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0533.933.6
TeCoABackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0522.422.3
TTCPre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Lightweight Defense, Input Modality=Vision Only2026.0310.621.39
C-TPT*Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations + Prompt Tuning, Input Modality=Vision + Text2026.0310.61.3
ET3Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Lightweight Defense, Input Modality=Vision Only2026.0310.322.04
CLIP (Robust)Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Baseline2026.0310.20.99
TPT*Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations + Prompt Tuning, Input Modality=Vision + Text2026.039.62
MTA*Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations, Input Modality=Vision Only2026.0392.5
APTBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.058.50.6
R-TPT + ET3Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations + Prompt Tuning, Input Modality=Vision + Text2026.036.213.27
Ensemble + ET3Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations, Input Modality=Vision Only2026.035.773.03
R-TPTPre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations + Prompt Tuning, Input Modality=Vision + Text2026.035.572.79
EnsemblePre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations, Input Modality=Vision Only2026.035.552.8