Urban Perception Prediction on Urban Perception Wealthy (test)
58.1Macro F1 ScoreGaze + Patch Transformer
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gaze + Patch TransformerRepresentation=Pretrained ViT patch, Gaze Integration=Explicit gaze token features2026.05 | 58.1 | 58 | |
| Gaze + Patch Transformer (Shuffled Patch Alignment)Ablation=Random alignment between gaze and patches2026.05 | 56.9 | 56.7 | |
| Patch Sequence TransformerRepresentation=Pretrained ViT patch, Gaze Integration=Sequence order only2026.05 | 56.8 | 56.9 | |
| Gaze + Patch Transformer (w/o Gaze)Ablation=No gaze input2026.05 | 56.7 | 56.7 | |
| Image-Only ViTRepresentation=Pretrained ViT patch2026.05 | 56.3 | 56.2 | |
| Gaze-weighted Patch PoolingRepresentation=Pretrained ViT patch, Fusion=Soft spatial prior reweighting2026.05 | 54.8 | 54.8 | |
| Gaze + AOI TransformerInput=Gaze + Semantic AOI2026.05 | 46.1 | 46.3 | |
| w/o GazeAblation=Removal of gaze component from Gaze + AOI Transformer2026.05 | 41.9 | 42.9 | |
| AOI Sequence TransformerInput=Semantic AOI sequence2026.05 | 41.8 | 43 | |
| AOI-Only ModelInput=Semantic AOI only2026.05 | 40.5 | 42.5 | |
| Shuffled AOI AlignmentAblation=Randomized alignment for Gaze + AOI Transformer2026.05 | 37.4 | 37.7 |