ReDepth Anything: Test-Time Depth Refinement via Self-Supervised Re-lighting

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Bhattarai, Ananta R., Rhodin, Helge
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912950152855552
author Bhattarai, Ananta R.
Rhodin, Helge
author_facet Bhattarai, Ananta R.
Rhodin, Helge
contents Monocular depth estimation remains challenging, as foundation models such as Depth Anything V2 (DA-V2) struggle with real-world images that are far from the training distribution. We introduce Re-Depth Anything, a test-time self-supervision framework that bridges this domain gap by fusing foundation models with the powerful priors of large-scale 2D diffusion models. Our method performs label-free refinement directly on the input image by re-lighting the predicted depth map and augmenting the input. This re-synthesis method replaces classical photometric reconstruction by leveraging shape from shading (SfS) cues in a new, generative context with Score Distillation Sampling (SDS). To prevent optimization collapse, our framework updates only intermediate embeddings and the decoder's weights, rather than optimizing the depth tensor directly or fine-tuning the full model. Across diverse benchmarks, Re-Depth Anything yields substantial gains in depth accuracy and realism over DA-V2, and applied on top of Depth Anything 3 (DA3) achieves state-of-the-art results, showcasing new avenues for self-supervision by geometric reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2512_17908
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ReDepth Anything: Test-Time Depth Refinement via Self-Supervised Re-lighting
Bhattarai, Ananta R.
Rhodin, Helge
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Monocular depth estimation remains challenging, as foundation models such as Depth Anything V2 (DA-V2) struggle with real-world images that are far from the training distribution. We introduce Re-Depth Anything, a test-time self-supervision framework that bridges this domain gap by fusing foundation models with the powerful priors of large-scale 2D diffusion models. Our method performs label-free refinement directly on the input image by re-lighting the predicted depth map and augmenting the input. This re-synthesis method replaces classical photometric reconstruction by leveraging shape from shading (SfS) cues in a new, generative context with Score Distillation Sampling (SDS). To prevent optimization collapse, our framework updates only intermediate embeddings and the decoder's weights, rather than optimizing the depth tensor directly or fine-tuning the full model. Across diverse benchmarks, Re-Depth Anything yields substantial gains in depth accuracy and realism over DA-V2, and applied on top of Depth Anything 3 (DA3) achieves state-of-the-art results, showcasing new avenues for self-supervision by geometric reasoning.
title ReDepth Anything: Test-Time Depth Refinement via Self-Supervised Re-lighting
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2512.17908