Iris: Bringing Real-World Priors into Diffusion Model for Monocular Depth Estimation

2026-03-17

AI Summary

Monocular depth estimation remains challenging in preserving fine details, generalizing to real-world scenes, and learning effectively from limited training data. This work proposes Iris, a novel framework that introduces Spectral Gated Distillation (SGD) and Spectral Gated Consistency (SGC) mechanisms, integrated via a two-stage Priors-to-Geometry Deterministic (PGD) scheduling strategy to effectively incorporate real-world priors into diffusion models. By synergistically leveraging both high- and low-frequency information, the method achieves a balanced trade-off between geometric fidelity and generalization during synthetic-to-real domain transfer. Consequently, Iris significantly improves depth estimation accuracy under data-scarce conditions and demonstrates superior generalization performance in diverse real-world scenarios.

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
• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

In this paper, we propose \textbf{Iris}, a deterministic framework for Monocular Depth Estimation (MDE) that integrates real-world priors into the diffusion model. Conventional feed-forward methods rely on massive training data, yet still miss details. Previous diffusion-based methods leverage rich generative priors yet struggle with synthetic-to-real domain transfer. Iris, in contrast, preserves fine details, generalizes strongly from synthetic to real scenes, and remains efficient with limited training data. To this end, we introduce a two-stage Priors-to-Geometry Deterministic (PGD) schedule: the prior stage uses Spectral-Gated Distillation (SGD) to transfer low-frequency real priors while leaving high-frequency details unconstrained, and the geometry stage applies Spectral-Gated Consistency (SGC) to enforce high-frequency fidelity while refining with synthetic ground truth. The two stages share weights and are executed with a high-to-low timestep schedule. Extensive experimental results confirm that Iris achieves significant improvements in MDE performance with strong in-the-wild generalization.

Problem

• Monocular Depth Estimation
• Diffusion Model
• Real-World Priors
• Domain Transfer
• Detail Preservation

Innovation

• diffusion model
• monocular depth estimation
• real-world priors
• spectral-gated distillation
• domain generalization