SelfOccFlow: Towards end-to-end self-supervised 3D Occupancy Flow prediction
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915823038234624 |
|---|---|
| author | Timoneda, Xavier Herb, Markus Duerr, Fabian Goehring, Daniel |
| author_facet | Timoneda, Xavier Herb, Markus Duerr, Fabian Goehring, Daniel |
| contents | Estimating 3D occupancy and motion at the vehicle's surroundings is essential for autonomous driving, enabling situational awareness in dynamic environments. Existing approaches jointly learn geometry and motion but rely on expensive 3D occupancy and flow annotations, velocity labels from bounding boxes, or pretrained optical flow models. We propose a self-supervised method for 3D occupancy flow estimation that eliminates the need for human-produced annotations or external flow supervision. Our method disentangles the scene into separate static and dynamic signed distance fields and learns motion implicitly through temporal aggregation. Additionally, we introduce a strong self-supervised flow cue derived from features' cosine similarities. We demonstrate the efficacy of our 3D occupancy flow method on SemanticKITTI, KITTI-MOT, and nuScenes. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_23894 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SelfOccFlow: Towards end-to-end self-supervised 3D Occupancy Flow prediction Timoneda, Xavier Herb, Markus Duerr, Fabian Goehring, Daniel Computer Vision and Pattern Recognition 68T40 I.2.9; I.2.10 Estimating 3D occupancy and motion at the vehicle's surroundings is essential for autonomous driving, enabling situational awareness in dynamic environments. Existing approaches jointly learn geometry and motion but rely on expensive 3D occupancy and flow annotations, velocity labels from bounding boxes, or pretrained optical flow models. We propose a self-supervised method for 3D occupancy flow estimation that eliminates the need for human-produced annotations or external flow supervision. Our method disentangles the scene into separate static and dynamic signed distance fields and learns motion implicitly through temporal aggregation. Additionally, we introduce a strong self-supervised flow cue derived from features' cosine similarities. We demonstrate the efficacy of our 3D occupancy flow method on SemanticKITTI, KITTI-MOT, and nuScenes. |
| title | SelfOccFlow: Towards end-to-end self-supervised 3D Occupancy Flow prediction |
| topic | Computer Vision and Pattern Recognition 68T40 I.2.9; I.2.10 |
| url | https://arxiv.org/abs/2602.23894 |