BokehDepth: Enhancing Monocular Depth Estimation through Bokeh Generation

2025-12-13

AI Summary

Monocular depth estimation suffers from degraded accuracy in weak-texture, long-range, and geometrically ambiguous regions—precisely where defocus (bokeh) cues are most discriminative; moreover, noisy depth maps severely impair bokeh rendering quality. To address this, we propose an unsupervised two-stage framework that, for the first time, models defocus as a physics-driven geometric prior to decouple and enhance depth estimation. First, we design a physics-guided, controllable bokeh generator leveraging a pre-trained image-editing backbone for high-fidelity defocus synthesis. Second, we introduce a lightweight defocus-aware feature aggregation module enabling cross-modal fusion between defocus and depth features. Our method is plug-and-play compatible with mainstream monocular depth encoders. Extensive experiments demonstrate significant improvements in depth metric accuracy and robustness across multiple benchmarks, while simultaneously producing more natural and photorealistic bokeh effects.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Bokeh and monocular depth estimation are tightly coupled through the same lens imaging geometry, yet current methods exploit this connection in incomplete ways. High-quality bokeh rendering pipelines typically depend on noisy depth maps, which amplify estimation errors into visible artifacts, while modern monocular metric depth models still struggle on weakly textured, distant and geometrically ambiguous regions where defocus cues are most informative. We introduce BokehDepth, a two-stage framework that decouples bokeh synthesis from depth prediction and treats defocus as an auxiliary supervision-free geometric cue. In Stage-1, a physically guided controllable bokeh generator, built on a powerful pretrained image editing backbone, produces depth-free bokeh stacks with calibrated bokeh strength from a single sharp input. In Stage-2, a lightweight defocus-aware aggregation module plugs into existing monocular depth encoders, fuses features along the defocus dimension, and exposes stable depth-sensitive variations while leaving downstream decoder unchanged. Across challenging benchmarks, BokehDepth improves visual fidelity over depth-map-based bokeh baselines and consistently boosts the metric accuracy and robustness of strong monocular depth foundation models.

Problem

• Enhances monocular depth estimation using bokeh generation as supervision-free cue
• Decouples bokeh synthesis from depth prediction to reduce visible artifacts
• Improves depth accuracy in weakly textured and ambiguous image regions

Innovation

• Two-stage framework decouples bokeh synthesis from depth prediction
• Controllable bokeh generator produces depth-free bokeh stacks from sharp input
• Defocus-aware aggregation module fuses features to enhance depth estimation