Communication Dans Un Congrès Année : 2025

Towards Sharper Object Boundaries in Self-Supervised Depth Estimation

Résumé

Accurate monocular depth estimation is crucial for 3D scene understanding, but existing methods often blur depth at object boundaries, introducing spurious intermediate 3D points. While achieving sharp edges usually requires very fine-grained supervision, our method produces crisp depth discontinuities using only self-supervision. Specifically, we model per-pixel depth as a mixture distribution, capturing multiple plausible depths and shifting uncertainty from direct regression to the mixture weights. This formulation integrates seamlessly into existing pipelines via variance-aware loss functions and uncertainty propagation. Extensive evaluations on KITTI and VKITTIv2 show that our method achieves up to 35\% higher boundary sharpness and improves point cloud quality compared to state-of-the-art baselines.

Fichier principal
Vignette du fichier
tsob.pdf (1.55 Mo) Télécharger le fichier
tsob_supp.pdf (953.05 Ko) Télécharger le fichier

Dates et versions

hal-05229697 , version 1 (29-08-2025)
hal-05229697 , version 2 (17-11-2025)

Licence

Identifiants

  • HAL Id : hal-05229697 , version 1

Citer

Aurélien Cecille, Stefan Duffner, Franck Davoine, Rémi Agier, Thibault Neveu. Towards Sharper Object Boundaries in Self-Supervised Depth Estimation. BMVC, BMVA, Nov 2025, Sheffield, United Kingdom. ⟨hal-05229697v1⟩
1129 Consultations
542 Téléchargements

Partager

  • More