Semantic Segmentation on OctoSense day (test)
41.1mIoULate-fusion MAE
Evaluation results
| Method | Links | |
|---|---|---|
| Late-fusion MAEFusion=late2026.06 | 41.1 | |
| DINOv2Encoder=DINOv22026.06 | 41 | |
| V-JEPA 2.1 (stereo)*Encoder=V-JEPA 2.1, Input=stereo2026.06 | 40.5 | |
| Late-fusion MAE (w/ full attention)Attention=full2026.06 | 40.5 | |
| Late-fusion MAE (w/o EV)Removed Modality=EV2026.06 | 40.3 | |
| V-JEPA 2.1Encoder=V-JEPA 2.12026.06 | 40.2 | |
| Late-fusion MAE (w/o IMU)Removed Modality=IMU2026.06 | 40.2 | |
| Perception Enc.Encoder=Perception Enc.2026.06 | 38.8 | |
| DINOv3Encoder=DINOv32026.06 | 38.2 | |
| Early-fusion MAEFusion=early2026.06 | 37.9 | |
| SigLIP 2Encoder=SigLIP 22026.06 | 37.7 | |
| Late-fusion MAE (w/o RGB)Removed Modality=RGB2026.06 | 36.6 | |
| Late-fusion MAE (w/o LiDAR)Removed Modality=LiDAR2026.06 | 35.1 | |
| RGB-only video MAEModality=RGB, Type=video2026.06 | 32.7 | |
| Late-fusion MAE (w/ only RGB)Modality=only RGB2026.06 | 31 | |
| RGB-only image MAEModality=RGB, Type=image2026.06 | 29.6 |