Object Counting on Clevr Count (test)
90.5AccuracyMAE
Evaluation results
| Method | Links | |
|---|---|---|
| MAEArch.=ViT-H/14, Evaluation protocol=Linear-probe, Pre-training data augmentation strategy=Methods without view data augmentations2023.01 | 90.5 | |
| DSeq-C-JEPABackbone=ViT-L/16, Evaluation Protocol=Linear probe on frozen features2025.11 | 87.1 | |
| C-JEPABackbone=ViT-L/16, Evaluation Protocol=Linear probe on frozen features2025.11 | 86.8 | |
| I-JEPAArch.=ViT-H/14, Evaluation protocol=Linear-probe, Pre-training data augmentation strategy=Methods without view data augmentations2023.01 | 86.7 | |
| DINOArch.=ViT-B/8, Evaluation protocol=Linear-probe, Pre-training data augmentation strategy=Methods using extra data augmentations2023.01 | 86.6 | |
| DSeq-JEPABackbone=ViT-L/16, Evaluation Protocol=Linear probe on frozen features2025.11 | 86.4 | |
| iBOTArch.=ViT-L/16, Evaluation protocol=Linear-probe, Pre-training data augmentation strategy=Methods using extra data augmentations2023.01 | 85.7 | |
| I-JEPABackbone=ViT-L/16, Evaluation Protocol=Linear probe on frozen features2025.11 | 85.6 | |
| data2vecArch.=ViT-L/16, Evaluation protocol=Linear-probe, Pre-training data augmentation strategy=Methods without view data augmentations2023.01 | 85.3 |