Reproduce “Boosting Monocular Metric Depth Estimation via Bokeh Rendering” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hangwei Zhang, Armando Fortes, Tianyi Wei, Xingang Pan

BokehDepth uses a calibrated stack of lens-inspired blurred versions of one image to help a metric-depth model infer distance, comparing features across blur strengths rather than deriving depth directly from blur. After fine-tuning on Hypersim, UniDepthV2’s IBims-1 zero-shot evaluation accuracy rises from 0.945 to 0.978, while AbsRel falls from 0.082 to 0.039. Gains require calibrated blur; evaluation is single-image.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.