Multi-label Classification on MS-COCO 2014
36.58Precision @ 3UCAT
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| UCATBackbone=CLIP-B/32, Evaluation Mode=Zero-shot, Evaluation Attack=CW-100, Training Attack=10-step PGD (ε = 2/255), Training Dataset=TinyImageNet2025.12 | 36.58 | 37.52 | 37.04 | 28.27 | 48.32 | 35.67 | 37.6 | |
| FAREBackbone=CLIP-B/32, Evaluation Mode=Zero-shot, Evaluation Attack=CW-100, Training Attack=10-step PGD (ε = 2/255), Training Dataset=TinyImageNet2025.12 | 33.45 | 34.3 | 33.86 | 26.04 | 44.49 | 32.84 | 29.18 | |
| TGA-ZSRBackbone=CLIP-B/32, Evaluation Mode=Zero-shot, Evaluation Attack=CW-100, Training Attack=10-step PGD (ε = 2/255), Training Dataset=TinyImageNet2025.12 | 32.95 | 33.79 | 33.36 | 24.58 | 41.99 | 31 | 38.23 | |
| PMG-AFTBackbone=CLIP-B/32, Evaluation Mode=Zero-shot, Evaluation Attack=CW-100, Training Attack=10-step PGD (ε = 2/255), Training Dataset=TinyImageNet2025.12 | 32.32 | 33.15 | 32.72 | 25.4 | 43.4 | 32.04 | 29.75 | |
| TeCoABackbone=CLIP-B/32, Evaluation Mode=Zero-shot, Evaluation Attack=CW-100, Training Attack=10-step PGD (ε = 2/255), Training Dataset=TinyImageNet2025.12 | 30.23 | 30.99 | 30.6 | 22.67 | 38.73 | 28.59 | 37.32 | |
| CLIPBackbone=CLIP-B/32, Evaluation Mode=Zero-shot, Evaluation Attack=CW-100, Training Attack=10-step PGD (ε = 2/255), Training Dataset=TinyImageNet2025.12 | 17.72 | 25.21 | 25.85 | 25.52 | 20.15 | 34.44 | 25.42 |