Optical Flow Estimation on OctoSense day (test)
1.96EPE (px)Late-fusion MAE (w/ full attention)
Evaluation Results
| Method | Links | |
|---|---|---|
| Late-fusion MAE (w/ full attention)Attention=full2026.06 | 1.96 | |
| Late-fusion MAEFusion=late2026.06 | 1.97 | |
| Late-fusion MAE (w/o EV)Removed Modality=EV2026.06 | 2 | |
| Early-fusion MAEFusion=early2026.06 | 2.01 | |
| Late-fusion MAE (w/o RGB)Removed Modality=RGB2026.06 | 2.02 | |
| Late-fusion MAE (w/o IMU)Removed Modality=IMU2026.06 | 2.05 | |
| Late-fusion MAE (w/o LiDAR)Removed Modality=LiDAR2026.06 | 7.19 | |
| V-JEPA 2.1 (stereo)*Encoder=V-JEPA 2.1, Input=stereo2026.06 | 9.03 | |
| V-JEPA 2.1Encoder=V-JEPA 2.12026.06 | 9.13 | |
| RGB-only video MAEModality=RGB, Type=video2026.06 | 10.11 | |
| Late-fusion MAE (w/ only RGB)Modality=only RGB2026.06 | 11.31 | |
| DINOv3Encoder=DINOv32026.06 | 19.43 | |
| DINOv2Encoder=DINOv22026.06 | 19.46 | |
| Perception Enc.Encoder=Perception Enc.2026.06 | 19.9 | |
| SigLIP 2Encoder=SigLIP 22026.06 | 20.22 | |
| RGB-only image MAEModality=RGB, Type=image2026.06 | 20.88 |