Semantic Segmentation on Moonstone Mare (Downstream)
93.6mIoUMG-MAE
Evaluation Results
| Method | Links | |
|---|---|---|
| MG-MAEPre-training=Lunar, Evaluation Protocol=Finetune2026.07 | 93.6 | |
| MG-MAEPre-training=Lunar, Evaluation Protocol=Linear2026.07 | 92.8 | |
| Vanilla MAEPre-training=Lunar, Details=Per-channel tokenizers without modality grouping2026.07 | 91.2 | |
| ScratchPre-training=None2026.07 | 90.1 | |
| ImageNet ViT-BPre-training=ImageNet, Evaluation Protocol=Linear2026.07 | 86.7 |