Saved in:
Bibliographic Details
Main Authors: Schmid, Jakob, Jahedi, Azin, Senn, Noah Berenguel, Bruhn, Andrés
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.01443
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909632435322880
author Schmid, Jakob
Jahedi, Azin
Senn, Noah Berenguel
Bruhn, Andrés
author_facet Schmid, Jakob
Jahedi, Azin
Senn, Noah Berenguel
Bruhn, Andrés
contents Although multi-scale concepts have recently proven useful for recurrent network architectures in the field of optical flow and stereo, they have not been considered for image-based scene flow so far. Hence, based on a single-scale recurrent scene flow backbone, we develop a multi-scale approach that generalizes successful hierarchical ideas from optical flow to image-based scene flow. By considering suitable concepts for the feature and the context encoder, the overall coarse-to-fine framework and the training loss, we succeed to design a scene flow approach that outperforms the current state of the art on KITTI and Spring by 8.7%(3.89 vs. 4.26) and 65.8% (9.13 vs. 26.71), respectively. Our code is available at https://github.com/cv-stuttgart/MS-RAFT-3D.
format Preprint
id arxiv_https___arxiv_org_abs_2506_01443
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MS-RAFT-3D: A Multi-Scale Architecture for Recurrent Image-Based Scene Flow
Schmid, Jakob
Jahedi, Azin
Senn, Noah Berenguel
Bruhn, Andrés
Computer Vision and Pattern Recognition
Although multi-scale concepts have recently proven useful for recurrent network architectures in the field of optical flow and stereo, they have not been considered for image-based scene flow so far. Hence, based on a single-scale recurrent scene flow backbone, we develop a multi-scale approach that generalizes successful hierarchical ideas from optical flow to image-based scene flow. By considering suitable concepts for the feature and the context encoder, the overall coarse-to-fine framework and the training loss, we succeed to design a scene flow approach that outperforms the current state of the art on KITTI and Spring by 8.7%(3.89 vs. 4.26) and 65.8% (9.13 vs. 26.71), respectively. Our code is available at https://github.com/cv-stuttgart/MS-RAFT-3D.
title MS-RAFT-3D: A Multi-Scale Architecture for Recurrent Image-Based Scene Flow
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.01443