Geometric Distillation from Rectified Stereo: Leveraging Epipolar Cues for Monocular Depth

2026-07-16

AI Summary

This work addresses the metric scale ambiguity inherent in monocular depth estimation across diverse scenes, which often yields geometrically plausible yet scale-inconsistent predictions. To resolve this, the authors propose Epipolar Distillation—a novel approach that distills scale-aware geometric priors from a multi-view model while preserving epipolar attention patterns through rectified stereo tokens. This enables the monocular model to achieve metrically aligned depth predictions at inference time without requiring multi-view inputs. The method seamlessly integrates into Vision Transformer (ViT) architectures such as UniDepthV2 and DepthPro, significantly enhancing zero-shot metric depth estimation performance. Extensive experiments on challenging benchmarks including ETH3D and DIODE demonstrate superior scale consistency and cross-dataset generalization capabilities.

Abstract

Monocular depth foundation models have demonstrated remarkable generalization capabilities across diverse environments. However, they continue to struggle with metric depth estimation in diverse environments. This limitation stems from the inherent scale ambiguity of single-view inference, leading to misaligned scale predictions even when the relative geometry is accurate. Conversely, recent multi-view foundation models leverage cross-view cues to learn robust scene-level geometry and consistent scale. Yet, these benefits typically vanish during single-image inference, as the absence of explicit geometric constraints causes performance to degrade. To bridge this gap, we propose a novel framework that transfers the scale-aware geometric priors of multi-view models into monocular depth foundation models. Specifically, we introduce an Epipolar Distillation (EpiDistill), an approach utilizing Rectified Stereo Tokens, which enables the single-view prediction model to retain epipolar attention patterns and maintain geometric consistency without requiring multi-view inputs at inference. Experimental results demonstrate that our method significantly improves zero-shot metric depth estimation, particularly on challenging datasets like ETH3D and DIODE where scale alignment is critical. Furthermore, our approach is model-agnostic, consistently boosting the performance of state-of-the-art ViT-based models, including UniDepthV2 and DepthPro.

Problem

• monocular depth estimation
• scale ambiguity
• metric depth
• geometric consistency
• epipolar constraints

Innovation

• Epipolar Distillation
• Monocular Depth Estimation
• Geometric Priors
• Rectified Stereo Tokens
• Metric Depth