List your company
ResearchFormerly WizWand

Fine-grained Classification on Pets

93.7AccuracyMTA
3.32426.78750.2573.713Nov 25, 2022Jun 27, 2023Jan 28, 2024Aug 30, 2024Apr 1, 2025Nov 2, 2025Jun 5, 2026
Updated 4mo ago

Evaluation results

MethodLinks
MTABackbone=ViT-L/14, Epsilon=4/2552026.0593.764.9
R-TPTBackbone=ViT-L/14, Epsilon=4/2552026.0593.772.9
R-TPTCategory=Tuning-based, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0693.772.9
MTACategory=Tuning-free, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0693.764.9
MACCategory=Tuning-free, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0693.687
AGCBackbone=ViT-L/14, Epsilon=4/2552026.0593.498.5
CLIPBackbone=ViT-L/14, Epsilon=4/2552026.0593.10
TTPBackbone=ViT-L/14, Epsilon=4/2552026.0593.176.3
CLIPCategory=Baseline, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0693.10
EnsembleBackbone=ViT-L/14, Epsilon=4/2552026.0592.577.6
TTCBackbone=ViT-L/14, Epsilon=4/2552026.0592.27.6
TTCCategory=Tuning-free, Backbone=CLIP-ViT-L/14, Attack Protocol=PGD-100, eps=4.02026.0692.26.4
CLIPBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0588.30
TTPBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0588.364.7
CLIPmode=Zero-shot2026.0288.25—
C-TPTmode=Test-Time Adaptation2026.0288.14—
MTABackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.058851.8
MTAmode=Test-Time Adaptation2026.0287.9—
TPSmode=Test-Time Adaptation2026.0287.44—
ZEROmode=Test-Time Adaptation2026.0287.33—
TPTmode=Test-Time Adaptation2026.0287.22—
TTLmode=Test-Time Adaptation2026.0287.22—
R-TPTBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0587.260.2
AGCBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.058796.1
RLCFmode=Test-Time Adaptation2026.0286.97—
R-TPTmode=Test-Time Adaptation2026.0286.73—
Fair Context Learning (FCL)mode=Test-Time Adaptation2026.0286.54—
CLIPBackbone=CLIP2022.1185.4—
EnsembleBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.058565.9
R-TPTBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0584.674.2
AGCBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.058485.1
CLIPBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0583.60
TTPBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0583.570.3
TTCBackbone=ViT-B/16, Pre-train=CLIP, Epsilon (ϵ)=4/255, Attack Protocol=PGD-1002026.0582.310.4
APT+TeCoABackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0579.379
TeCoABackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.057675.8
GIF-SDBackbone=ResNet-50, Expansion Ratio=30x2022.1173.4—
CLIP (Robust)Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Baseline2026.0366.8815.94
ET3Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Lightweight Defense, Input Modality=Vision Only2026.0366.8627.15
GIF-DALLEBackbone=ResNet-50, Expansion Ratio=30x2022.1166.4—
MTA*Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations, Input Modality=Vision Only2026.0366.231.2
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 + Text2026.0366.119.5
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 + Text2026.0365.227.4
TTCPre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Lightweight Defense, Input Modality=Vision Only2026.036521.1
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 + Text2026.0363.8645.71
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 + Text2026.0363.7541.43
Ensemble + ET3Pre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations, Input Modality=Vision Only2026.0362.0343.17
DALL-E2Backbone=ResNet-50, Expansion Ratio=30x2022.1161.7—
EnsemblePre-trained Model=TeCoA CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4/255), Defense Category=Multiple Augmentations, Input Modality=Vision Only2026.0359.9638.35
SDBackbone=ResNet-50, Expansion Ratio=30x2022.1157.9—
GIF-MAEBackbone=ResNet-50, Expansion Ratio=30x2022.1152.4—
RandAugmentBackbone=ResNet-50, Expansion Ratio=30x2022.1148—
MAEBackbone=ResNet-50, Expansion Ratio=30x2022.1139.9—
CutoutBackbone=ResNet-50, Expansion Ratio=30x2022.1138.7—
GridMaskBackbone=ResNet-50, Expansion Ratio=30x2022.1137.6—
APTBackbone=ResNet50, Attack Type=PGD, Epsilon (ε)=1.02026.0531.93.8
Distillation of CLIPBackbone=ResNet-502022.1111.1—
OriginalBackbone=ResNet-50, Expansion Ratio=1x2022.116.8—