Large-Scale Light Field Synthesis from Videos Enables Geometrically Consistent Bokeh Editing

Haoming Cai Zhoutong Zhang Christopher A. Metzler Shumian Xin
European Conference on Computer Vision (ECCV) — 2026
LightFieldBokeh teaser

Realistic post-capture bokeh (defocus) editing requires geometric scene understanding that single images lack. We synthesize large-scale light fields from ordinary videos and use them to build an image bokeh dataset that enables joint focus and depth-of-field editing on any image. The resulting method produces geometrically consistent bokeh with realistic, controllable defocus.

@article{ Cai2026LightFieldBokeh,
  author    = { Cai, Haoming and Zhang, Zhoutong and Metzler, Christopher A. and Xin, Shumian },
  title     = { Large-Scale Light Field Synthesis from Videos Enables Geometrically Consistent Bokeh Editing },
  journal   = { European Conference on Computer Vision (ECCV) },
  year      = { 2026 },
}