Zur Kurzanzeige

Boosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Model

dc.contributor.authorEndres, Jannik
dc.contributor.authorHahn, Oliver
dc.contributor.authorCorbière, Charles
dc.contributor.authorSchaub-Meyer, Simone
dc.contributor.authorRoth, Stefan
dc.contributor.authorAlahi, Alexandre
dc.date.accessioned2025-04-10T14:03:25Z
dc.date.available2025-04-10T14:03:25Z
dc.date.issued2025-03
dc.identifier.urihttps://tudatalib.ulb.tu-darmstadt.de/handle/tudatalib/4557
dc.descriptionOmnidirectional depth perception is essential for mobile robotics applications that require scene understanding across a full 360° field of view. Camera-based setups offer a cost-effective option by using stereo depth estimation to generate dense, high-resolution depth maps without relying on expensive active sensing. However, existing omnidirectional stereo matching approaches achieve only limited depth accuracy across diverse environments, depth ranges, and lighting conditions, due to the scarcity of real-world data. We present DFI-OmniStereo, a novel omnidirectional stereo matching method that leverages a large-scale pre-trained foundation model for relative monocular depth estimation within an iterative optimization-based stereo matching architecture. We introduce a dedicated two-stage training strategy to utilize the relative monocular depth features for our omnidirectional stereo matching before scale-invariant fine-tuning. DFI-OmniStereo achieves state-of-the-art results on the real-world Helvipad dataset, reducing disparity MAE by approximately 16% compared to the previous best omnidirectional stereo method.de_DE
dc.language.isoende_DE
dc.relationIsDescribedBy;arXiv;2503.23502
dc.rightsApache License 2.0
dc.rights.urihttps://www.apache.org/licenses/LICENSE-2.0
dc.subjectDepth Predictionde_DE
dc.subjectOmnidirectional depth perceptionde_DE
dc.subjectStereo Matchingde_DE
dc.subjectdeep learningde_DE
dc.subjectfoundation modelde_DE
dc.subject.classification4.43-04 Künstliche Intelligenz und Maschinelle Lernverfahrende_DE
dc.subject.classification4.43-05 Bild- und Sprachverarbeitung, Computergraphik und Visualisierung, Human Computer Interaction, Ubiquitous und Wearable Computingde_DE
dc.subject.ddc004
dc.titleBoosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Modelde_DE
dc.typeSoftwarede_DE
tud.projectEC/H2020 | 866008 | REDde_DE
tud.projectHMWK | III L6-519/03/05.001-(0016) | emergenCity - TP Rothde_DE
tud.projectHMWK | 500/10.001-(00012) | TAM - TP Rothde_DE
tud.unitTUDa


Dateien zu dieser Ressource

No Thumbnail [100%x60]
Name:
dfi_omnistereo_code.zip
Größe:
2.960MB
Format:
application/zip
Beschreibung:
Training and Inference Code ...
Anzahl der Dateien
No Thumbnail [100%x60]
Name:
dfi_omnistereo_helvipad.pth
Größe:
412.3MB
Format:
Unbekannt
Beschreibung:
Model parameters
Anzahl der Dateien
No Thumbnail [100%x60]
Name:
license_Apache-License.txt
Größe:
11.09KB
Format:
Text
Anzahl der Dateien

Der Datensatz erscheint in:

Zur Kurzanzeige

Apache License 2.0
Solange nicht anders angezeigt, wird die Lizenz wie folgt beschrieben: Apache License 2.0