Fine-grained classification on Pets (test)
85.9AccuracyMTA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MTACategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 85.9 | — | 47.8 | |
| CLIPCategory=Baseline, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 85.3 | — | 0 | |
| MACCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 85.2 | — | 68 | |
| R-TPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 84.5 | — | 56.5 | |
| CLIPBackbone=CLIP-ResNet50, Epsilon=1.02026.06 | 83.6 | — | 0.1 | |
| TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 83.6 | — | 0.1 | |
| SS-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 83.5 | — | 72.8 | |
| R-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 82.6 | — | 68.8 | |
| TAPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 82.5 | — | 14.6 | |
| EnsembleBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06 | 82.4 | — | 59.7 | |
| TTCCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06 | 81.9 | — | 11.2 | |
| TTCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06 | 76 | — | 10.7 | |
| DOCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06 | 75.3 | — | 10.3 | |
| GRPOAnnotation Noise Level=0%, Number of shots=4-shot2025.11 | 70.7 | — | — | |
| GRPO w. Min.Annotation Noise Level=0%, Number of shots=4-shot2025.11 | 70.4 | — | — | |
| GRPO w. Two.Annotation Noise Level=0%, Number of shots=4-shot2025.11 | 70 | — | — | |
| GRPO w. Max.Annotation Noise Level=0%, Number of shots=4-shot2025.11 | 69.8 | — | — | |
| GRPO w. Two.Annotation Noise Level=20%, Number of shots=4-shot2025.11 | 69.7 | — | — | |
| GRPO w. Min.Annotation Noise Level=20%, Number of shots=4-shot2025.11 | 69.5 | — | — | |
| GRPO w. Two.Annotation Noise Level=40%, Number of shots=4-shot2025.11 | 69 | — | — | |
| GRPO w. Max.Annotation Noise Level=20%, Number of shots=4-shot2025.11 | 68.9 | — | — | |
| GRPOAnnotation Noise Level=40%, Number of shots=4-shot2025.11 | 68.8 | — | — | |
| GRPOAnnotation Noise Level=20%, Number of shots=4-shot2025.11 | 68.8 | — | — | |
| GRPO w. Min.Annotation Noise Level=40%, Number of shots=4-shot2025.11 | 68.7 | — | — | |
| GRPO w. Min.Annotation Noise Level=50%, Number of shots=4-shot2025.11 | 68.6 | — | — | |
| GRPOAnnotation Noise Level=50%, Number of shots=4-shot2025.11 | 68.5 | — | — | |
| GRPO w. Max.Annotation Noise Level=40%, Number of shots=4-shot2025.11 | 68.5 | — | — | |
| GRPO w. Two.Annotation Noise Level=50%, Number of shots=4-shot2025.11 | 68.4 | — | — | |
| GRPO w. Max.Annotation Noise Level=50%, Number of shots=4-shot2025.11 | 67.8 | — | — | |
| GRPO w. Max.Annotation Noise Level=60%, Number of shots=4-shot2025.11 | 67.5 | — | — | |
| GRPO w. Two.Annotation Noise Level=60%, Number of shots=4-shot2025.11 | 67.5 | — | — | |
| GRPOAnnotation Noise Level=60%, Number of shots=4-shot2025.11 | 67.3 | — | — | |
| GRPO w. Min.Annotation Noise Level=60%, Number of shots=4-shot2025.11 | 66.9 | — | — | |
| GRPO w. Two.Annotation Noise Level=80%, Number of shots=4-shot2025.11 | 65.5 | — | — | |
| GRPOAnnotation Noise Level=80%, Number of shots=4-shot2025.11 | 64.7 | — | — | |
| GRPO w. Min.Annotation Noise Level=80%, Number of shots=4-shot2025.11 | 64.6 | — | — | |
| GRPO w. Max.Annotation Noise Level=80%, Number of shots=4-shot2025.11 | 64.5 | — | — | |
| GRPO w. Min.Annotation Noise Level=100%, Number of shots=4-shot2025.11 | 59.3 | — | — | |
| Base ModelAnnotation Noise Level=Base, Number of shots=4-shot2025.11 | 59.2 | — | — | |
| GRPOAnnotation Noise Level=100%, Number of shots=4-shot2025.11 | 54.7 | — | — | |
| GRPO w. Two.Annotation Noise Level=100%, Number of shots=4-shot2025.11 | 54.3 | — | — | |
| GRPO w. Max.Annotation Noise Level=100%, Number of shots=4-shot2025.11 | 51 | — | — |