Scanpath Prediction on Visual Search Dataset Clean Inputs
86.6BBox Hit RatioClean Model
Evaluation results
| Method | Links | |
|---|---|---|
| Clean ModelTrigger=None2026.04 | 86.6 | |
| Input-aware spatial attackTrigger modality=Vision, Poisoning ratio=10%2026.04 | 83.5 | |
| Input-aware spatial attackTrigger modality=Language, Poisoning ratio=5%2026.04 | 82.2 | |
| Input-aware spatial attackTrigger modality=Vision–Language, Poisoning ratio=10%2026.04 | 82 | |
| Input-aware spatial attackTrigger modality=Language, Poisoning ratio=2.5%2026.04 | 81.5 | |
| Input-aware spatial attackTrigger modality=Language, Poisoning ratio=10%2026.04 | 81 | |
| Input-aware spatial attackTrigger modality=Vision–Language, Poisoning ratio=5%2026.04 | 80.9 | |
| Input-aware spatial attackTrigger modality=Vision–Language, Poisoning ratio=2.5%2026.04 | 79.7 | |
| Input-aware spatial attackTrigger modality=Vision, Poisoning ratio=5%2026.04 | 79.6 | |
| Input-aware spatial attackTrigger modality=Vision, Poisoning ratio=2.5%2026.04 | 78.8 |