GeoDiff: Geometry-Guided Diffusion for Metric Depth Estimation

2025-10-21

AI Summary

Monocular metric depth estimation suffers from inherent scale ambiguity, particularly failing in scenes lacking geometric priors—e.g., transparent or specular surfaces. To address this, we propose a training-free inverse problem framework leveraging diffusion models: it employs a pre-trained latent diffusion model (LDM) as a generative prior, conditions on the input RGB image, and—crucially—introduces differentiable stereo-geometric constraints (e.g., epipolar consistency and disparity-depth mapping) as regularization terms to jointly optimize absolute depth maps. By eliminating reliance on supervised fine-tuning, our method is fully plug-and-play. It achieves state-of-the-art performance across diverse indoor and outdoor scenes, significantly improving depth accuracy and robustness on challenging surfaces. This work establishes a novel paradigm integrating generative modeling with geometric reasoning, enabling geometry-aware depth reconstruction without task-specific training.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

We introduce a novel framework for metric depth estimation that enhances pretrained diffusion-based monocular depth estimation (DB-MDE) models with stereo vision guidance. While existing DB-MDE methods excel at predicting relative depth, estimating absolute metric depth remains challenging due to scale ambiguities in single-image scenarios. To address this, we reframe depth estimation as an inverse problem, leveraging pretrained latent diffusion models (LDMs) conditioned on RGB images, combined with stereo-based geometric constraints, to learn scale and shift for accurate depth recovery. Our training-free solution seamlessly integrates into existing DB-MDE frameworks and generalizes across indoor, outdoor, and complex environments. Extensive experiments demonstrate that our approach matches or surpasses state-of-the-art methods, particularly in challenging scenarios involving translucent and specular surfaces, all without requiring retraining.

Problem

• Addresses metric depth estimation from single images
• Resolves scale ambiguities using stereo geometric constraints
• Enhances pretrained diffusion models without retraining

Innovation

• Geometry-guided diffusion for metric depth estimation
• Stereo vision constraints resolve scale ambiguities
• Training-free integration into existing depth frameworks