Image Classification on Oxford Flowers 102 (Accuracy and ECE)
99.61AccuracyFull Fine-tuning
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full Fine-tuningBackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 99.61 | — | — | |
| BaLoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 99.61 | — | — | |
| DoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 99.48 | — | — | |
| LoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | 99.44 | — | — | |
| APTBackbone=ViT-B/32, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Few-shot2026.06 | 84.4 | — | 47 | |
| DBDBackbone=ViT-B/16, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 73.6 | — | 97.5 | |
| CLIPBackbone=ViT-B/16, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 73 | — | 0 | |
| MaPLe+O-TPTBackbone=CLIP-ViT/B16, Prompt Initialization Source=MaPLe, Adaptation Framework=TPT, Adaptation Strategy=O-TPT2026.04 | 71.53 | 4.35 | — | |
| MaPLe+TPTBackbone=CLIP-ViT/B16, Prompt Initialization Source=MaPLe, Adaptation Framework=TPT, Adaptation Strategy=TPT2026.04 | 70.72 | 11.63 | — | |
| MaPLe+FPPBackbone=CLIP-ViT/B16, Prompt Initialization Source=MaPLe, Adaptation Framework=TPT, Adaptation Strategy=FPP2026.04 | 69.35 | 4.03 | — | |
| CoOp+C-TPTBackbone=CLIP-ViT/B16, Prompt Initialization Source=CoOp, Adaptation Framework=TPT, Adaptation Strategy=C-TPT2026.04 | 69 | 10.2 | — | |
| CoOp+TPTBackbone=CLIP-ViT/B16, Prompt Initialization Source=CoOp, Adaptation Framework=TPT, Adaptation Strategy=TPT2026.04 | 68.7 | 19.9 | — | |
| DBDBackbone=ViT-B/32, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 68.7 | — | 94.7 | |
| CoOp+O-TPTBackbone=CLIP-ViT/B16, Prompt Initialization Source=CoOp, Adaptation Framework=TPT, Adaptation Strategy=O-TPT2026.04 | 68.57 | 6.81 | — | |
| CoOp+FPPBackbone=CLIP-ViT/B16, Prompt Initialization Source=CoOp, Adaptation Framework=TPT, Adaptation Strategy=FPP2026.04 | 68.57 | 6 | — | |
| TTCBackbone=ViT-B/16, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 68.5 | — | 19.8 | |
| R-TPTBackbone=ViT-B/16, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 68.1 | — | 46.4 | |
| CLIPBackbone=ViT-B/32, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 67 | — | 0 | |
| CLIPBackbone=ResNet50, Attack Protocol=PGD-10, Epsilon (ε)=1/255, Evaluation Setting=Zero-shot2026.06 | 65.9 | — | 0 | |
| DBDBackbone=ResNet50, Attack Protocol=PGD-10, Epsilon (ε)=1/255, Evaluation Setting=Zero-shot2026.06 | 65.9 | — | 88.3 | |
| TTCBackbone=ViT-B/32, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 65.5 | — | 29.2 | |
| R-TPTBackbone=ViT-B/32, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 63.3 | — | 38.8 | |
| R-TPTBackbone=ResNet50, Attack Protocol=PGD-10, Epsilon (ε)=1/255, Evaluation Setting=Zero-shot2026.06 | 60.7 | — | 51 | |
| TeCoABackbone=ResNet50, Attack Protocol=PGD-10, Epsilon (ε)=1/255, Evaluation Setting=Zero-shot2026.06 | 32.6 | — | 17.3 | |
| CLIP-TeCoABackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 30.8 | — | 9 | |
| TeCoABackbone=ViT-B/32, Attack Protocol=PGD-100, Epsilon (ε)=4/255, Evaluation Setting=Zero-shot2026.06 | 30.1 | — | 9.5 | |
| C-TPTBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 29.4 | — | 10.7 | |
| MTABackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 29.1 | — | 14 | |
| TPTBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 27.9 | — | 12.3 | |
| EnsembleBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 26.6 | — | 16 | |
| R-TPTBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 26.6 | — | 16.5 | |
| T-VSSBackbone=TeCoA-pretrained CLIP-ViT-B/32, Epsilon (ϵ)=4/2552026.06 | 25.7 | — | 16.4 | |
| BaLoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | — | 0.63 | — | |
| DoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | — | 0.7 | — | |
| LoRABackbone=ViT-L/16, Pre-trained=ImageNet12K2026.04 | — | 0.62 | — |