Research map / Image generation / Fast diffusion sampling
BokehDiff: Neural Lens Blur with One-Step Diffusion
Chengxuan Zhu, Qingnan Fan, Qi Zhang, Jinwei Chen, Huaqi Zhang, Chao Xu, Boxin Shi
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 |
|---|---|---|---|
| add adversarial loss (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | LPIPS: 0.3768 -> 0.3737 | Improved |
| add VGG perceptual loss (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | LPIPS: 0.4218 -> 0.3737 | Improved |
| add multi-scale edge loss (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | LPIPS: 0.3785 -> 0.3737 | Improved |
| train the VAE encoder instead of fixing it (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | PSNR: 24.266 -> 24.652 | Improved |
| add circle-of-confusion constraint in PISA (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | PSNR: 22.217 -> 24.652 | Improved |
| add energy-conserved normalization (SoftmaxQ) (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | LPIPS: 0.3800 -> 0.3737 | Improved |
| add self-occlusion in PISA (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | LPIPS: 0.3808 -> 0.3737 | Improved |
| use timestep T=499 instead of 249 (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | PSNR: 24.646 -> 24.652 | No clear effect |
| use timestep T=499 instead of 749 (ablation) | BokehDiff · EBB Val294 (exposure-aligned) | PSNR: 24.481 -> 24.652 | No clear effect |
| use BokehDiff instead of BokehMe | BokehMe · BLB Level 5 (synthetic) | PSNR: 38.802 -> 36.798 | Hurt |
| use BokehDiff instead of BokehMe | BokehMe · BLB Level 5 (synthetic) | LPIPS: 0.1404 -> 0.0888 | Improved |
| use BokehDiff instead of BokehMe | BokehMe · EBB Val294 (exposure-aligned) | PSNR: 24.014 -> 24.652 | Improved |
Later papers that compared with this method
2 later accepted papers report BokehDiff as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- AnyBokeh: Physics-Guided Any-to-Any Bokeh Editing with Optical Fingerprint Transfer (NeurIPS 2026 · better on 4 of 4 shared results)
- LensStyle: Learning the Optical Aesthetics for Controllable Stylized Lens Effect Rendering (ECCV 2026 · better on 2 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Dr. Bokeh: DiffeRentiable Occlusion-aware Bokeh Rendering (CVPR 2024 · similarity 0.88)
- A Physics-Informed Blur Learning Framework for Imaging Systems (CVPR 2025 · similarity 0.86)
- BLURD: Benchmarking and Learning using a Unified Rendering and Diffusion Model (NeurIPS 2024 · similarity 0.86)
- Light Transport-aware Diffusion Posterior Sampling for Single-View Reconstruction of 3D Volumes (CVPR 2025 · similarity 0.85)
- LENS: Low-Frequency Eigen Noise Shaping for Efficient Diffusion Sampling (NeurIPS 2026 · similarity 0.85)
- AnyBokeh: Physics-Guided Any-to-Any Bokeh Editing with Optical Fingerprint Transfer (NeurIPS 2026 · 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.
