Gaze-following on VideoAttentionTarget (VAT) Consistent
0.959AUCHCLoRA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HCLoRABackbone=ViT-L2026.05 | 0.959 | 0.063 | 7.29 | |
| GazeLLE + FTBackbone=ViT-B, Adaptation protocol=Full fine-tuning2026.05 | 0.9538 | 0.0746 | 9.6 | |
| HCLoRABackbone=ViT-B2026.05 | 0.9534 | 0.0686 | 8.42 | |
| GazeLLEBackbone=ViT-B2026.05 | 0.9528 | 0.0693 | 9.21 | |
| GazeLLE + LoRABackbone=ViT-B, Adaptation protocol=LoRA2026.05 | 0.9477 | 0.0724 | 9.35 | |
| GazeLLE + VPTBackbone=ViT-B, Adaptation protocol=Visual Prompt Tuning2026.05 | 0.9455 | 0.0709 | 9.31 | |
| Sharingan2026.05 | 0.932 | 0.0741 | 9.41 | |
| Miao et al.2026.05 | 0.9295 | 0.0796 | 9.77 | |
| Chong et al.2026.05 | 0.8712 | 0.1237 | 15.56 |