Research map / 3D vision / Gaussian splatting
Large-Scale Light Field Synthesis from Videos Enables Geometrically Consistent Bokeh Editing
Haoming Cai, Zhoutong Zhang, Christopher Metzler, Shumian Xin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use light-field bokeh data from Vid2Bokeh instead of depth-based synthesis (BokehDiff) | bokeh data acquisition · Blender Direct Bokeh reference | PSNR: 28.70 -> 29.71 | Improved |
| use light-field bokeh data from Vid2Bokeh instead of depth-based synthesis (BokehDiff) | bokeh data acquisition · Blender Direct Bokeh reference | LPIPS: 0.141 -> 0.108 | Improved |
| use Vid2Bokeh instead of BokehDiff | bokeh data acquisition · Blender LF-based Bokeh reference | PSNR: 30.72 -> 33.37 | Improved |
| train editing model on Vid2Bokeh instead of DiffCamera | focus-DoF editing model · Refocusing near to far | PSNR: 28.32 -> 29.62 | Improved |
| train editing model on Vid2Bokeh instead of DiffCamera | focus-DoF editing model · Refocusing far to near | LPIPS: 0.201 -> 0.183 | Improved |
| our model vs DRBNet + BokehDiff | focus-DoF editing model · Refocusing near to far | SSIM: 0.834 -> 0.907 | Improved |
| our model vs Bokehlicious on EBB | bokeh rendering · EBB val | PSNR: 24.34 -> 24.02 | Hurt |
| our model vs Bokehlicious on EBB | bokeh rendering · EBB val | LPIPS: 0.182 -> 0.169 | Improved |
| our model vs Bokehlicious on Vid2Bokeh test | bokeh rendering · Vid2Bokeh test (avg 5 apertures) | PSNR: 27.42 -> 28.05 | Improved |
| our model vs Restormer on DPDD | defocus deblurring · DPDD | PSNR: 26.77 -> 26.70 | No clear effect |
| our model vs Restormer on Vid2Bokeh test | defocus deblurring · Vid2Bokeh test (avg 5 apertures) | PSNR: 27.81 -> 28.58 | Improved |
Closest papers by meaning
- BokehFlow: Depth-Free Controllable Bokeh Rendering via Flow Matching (AAAI 2026 · similarity 0.89)
- BokehCrafter: Taming Video Diffusion Models for Controllable Bokeh Rendering (AAAI 2026 · similarity 0.89)
- AnyBokeh: Physics-Guided Any-to-Any Bokeh Editing with Optical Fingerprint Transfer (NeurIPS 2026 · similarity 0.89)
- Any-to-Bokeh: Arbitrary-Subject Video Refocusing with Video Diffusion Model (ICLR 2026 · similarity 0.89)
- Dr. Bokeh: DiffeRentiable Occlusion-aware Bokeh Rendering (CVPR 2024 · similarity 0.87)
- S2D-LFE: Sparse-to-Dense Light Field Event Generation (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
