Fine-grained classification on Caltech101 (test)
98.26AccuracyOracle ADAPT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Oracle ADAPTBP-free=true, Evaluation Protocol=Transductive2025.08 | 98.26 | — | — | — | — | |
| ADAPTBP-free=true, Evaluation Protocol=Transductive2025.08 | 95.46 | — | — | — | — | |
| ADAPTBP-free=true, Evaluation Protocol=Online2025.08 | 94.48 | — | — | — | — | |
| O-TPTVenue=CVPR 2025, Backbone=CLIP-ViT/B16, Framework=TPT, Prompt Template=a photo of a2026.04 | 93.95 | — | — | 3.8 | 17 | |
| TPTVenue=NeurIPS 2022, Backbone=CLIP-ViT/B16, Framework=TPT, Prompt Template=a photo of a2026.04 | 93.8 | — | — | 4.51 | 16 | |
| C-TPTVenue=ICLR 2024, Backbone=CLIP-ViT/B16, Framework=TPT, Prompt Template=a photo of a2026.04 | 93.6 | — | — | 4.24 | 22 | |
| FPPVenue=CVPR 2026, Backbone=CLIP-ViT/B16, Framework=TPT, Prompt Template=a photo of a2026.04 | 93.25 | — | — | 5.85 | 26 | |
| CLIPBP-free=N/A, Evaluation Protocol=Online2025.08 | 92.98 | — | — | — | — | |
| CLIPVenue=ICML 2021, Backbone=CLIP-ViT/B16, Framework=TPT, Prompt Template=a photo of a2026.04 | 92.9 | — | — | 5.5 | 25 | |
| MTACategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 91.4 | — | 69.1 | — | — | |
| CLIPCategory=Baseline, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 91.3 | — | 0.1 | — | — | |
| MACCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 90.9 | — | 80.6 | — | — | |
| R-TPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 90.6 | — | 78 | — | — | |
| TTCCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 90.5 | — | 26.3 | — | — | |
| TAPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 90.4 | — | 38.7 | — | — | |
| TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 87.1 | — | 6.8 | — | — | |
| CLIPBackbone=CLIP-ResNet50, Epsilon=1.02026.06 | 85.7 | — | 5.2 | — | — | |
| SS-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 85.6 | — | 81 | — | — | |
| EnsembleBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 83.6 | — | 67.9 | — | — | |
| R-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 83.5 | — | 69.3 | — | — | |
| TTCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06 | 82 | — | 47.4 | — | — | |
| DOCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06 | 81.7 | — | 46.7 | — | — |