Research map / 3D vision / NeRF and radiance fields
RaFE: Generative Radiance Fields Restoration
Zhongkai Wu, Ziyu Wan, Jing Zhang, Jing Liao, Dong Xu
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 |
|---|---|---|---|
| replace per-frame NeRF fitting on restored images with a GAN-based generative two-level tri-plane NeRF | RaFE · Blender 4x super-resolution | LPIPS: 0.083 -> 0.062 | Improved |
| use generative modeling instead of per-frame fitting | RaFE · Blender 4x super-resolution | PSNR: 27.39 -> 24.99 | Hurt |
| use generative modeling instead of per-frame fitting | RaFE · LLFF 4x super-resolution | LPIPS: 0.261 -> 0.197 | Improved |
| RaFE vs per-frame baseline on deblurring | RaFE · LLFF deblurring (consistent blur) | LPIPS: 0.310 -> 0.252 | Improved |
| RaFE vs per-frame baseline on denoising | RaFE · LLFF denoising (Gain 8) | LPIPS: 0.281 -> 0.2561 | Improved |
| RaFE vs task-specific BAD-NeRF on pixel metric PSNR | RaFE · LLFF deblurring (consistent blur) | PSNR: 25.40 -> 22.93 | Hurt |
| NeRFLiX without ground-truth reference images vs RaFE on mixed degradation | NeRFLiX · Blender mixed degradation | LPIPS: 0.076 -> 0.107 | Hurt |
| add the residual tri-plane generator (vs directly optimizing NeRF with GAN and LPIPS loss) (ablation) | RaFE · Lego | image quality metrics | Improved |
| add the geometry (LPIPS) loss alongside GAN loss (ablation) | RaFE · Drums | PSNR | Improved |
| add GAN loss alongside perceptual loss (ablation) | RaFE · Drums | perceptual metrics | Improved |
| swap diffusion-based DiffBIR for non-diffusion SRFormer as the 2D restorer (ablation) | RaFE · Blender super-resolution | performance vs NeRF-Perframe | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- BAD-NeRF — BAD-NeRF: Bundle Adjusted Deblur Neural Radiance Fields (CVPR 2023) · better on 1 of 3 shared results
- Deblur-NeRF — Deblur-NeRF: Neural Radiance Fields From Blurry Images (CVPR 2022) · better on 1 of 1 shared result
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
- From NeRFLiX to NeRFLiX++: A General NeRF-Agnostic Restorer Paradigm (TPAMI 2023 · similarity 0.91)
- Neural Radiance Fields From Sparse RGB-D Images for High-Quality View Synthesis (TPAMI 2022 · similarity 0.91)
- NeRFLix: High-Quality Neural View Synthesis by Learning a Degradation-Driven Inter-Viewpoint MiXer (CVPR 2023 · similarity 0.91)
- Optimize the Unseen - Fast NeRF Cleanup with Free Space Prior (NeurIPS 2025 · similarity 0.89)
- ZeroRF: Fast Sparse View 360deg Reconstruction with Zero Pretraining (CVPR 2024 · similarity 0.89)
- Bright-NeRF: Brightening Neural Radiance Field with Color Restoration from Low-Light RAW Images (AAAI 2025 · similarity 0.88)
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.
