Object Recognition on YTVIS HQ
91.9Top-1 Accuracy (Class)RandSF.Q + xSSC
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RandSF.Q + xSSCEvaluation Protocol=freeze OCL + train MLP2026.05 | 91.9 | 97.9 | 52.9 | 9,183 | |
| SmoothSA + xSSCEvaluation Protocol=freeze OCL + train MLP2026.05 | 91.4 | 98.2 | 43.5 | 8,777 | |
| TSAEvaluation protocol=Two-layer MLP trained on frozen slot representations2026.06 | 91.4 | 98 | 50 | 7,843 | |
| RandSF.QNumber of slots=7, Evaluation protocol=+MLP, Backbone=DINO2 ViT-S/14, Input resolution=256×256 (224×224)2025.08 | 90.5 | 97.9 | 50.6 | 8,979 | |
| RandSF.QEvaluation Protocol=freeze OCL + train MLP2026.05 | 90.5 | 97.9 | 50.6 | 8,979 | |
| RandSF.Q+MLPEvaluation protocol=Two-layer MLP trained on frozen slot representations2026.06 | 90.5 | 97.9 | 50.6 | 8,979 | |
| SmoothSAEvaluation Protocol=freeze OCL + train MLP2026.05 | 90.4 | 97.6 | 42.6 | 8,957 | |
| SlotContrast + xSSCEvaluation Protocol=freeze OCL + train MLP2026.05 | 86.4 | 95.4 | 52.4 | 9,253 | |
| SlotContrastNumber of slots=7, Evaluation protocol=+MLP, Backbone=DINO2 ViT-S/14, Input resolution=256×256 (224×224)2025.08 | 85.8 | 95.8 | 51.5 | 9,249 | |
| SlotContrastEvaluation Protocol=freeze OCL + train MLP2026.05 | 85.8 | 95.8 | 51.5 | 9,249 | |
| SlotContrast+MLPEvaluation protocol=Two-layer MLP trained on frozen slot representations2026.06 | 85.8 | 95.8 | 51.5 | 9,249 |