Research map / 3D vision / NeRF and radiance fields
NeRF in the Dark: High Dynamic Range View Synthesis From Noisy Raw Images
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train NeRF directly on linear raw images rather than tonemapped LDR images | NeRF · Synthetic Lego with simulated noise | sRGB PSNR: 31.25 -> 32.11; 29.14 -> 31.94; 26.10 -> 31.59; 22.31 -> 30.94; 18.27 -> 29.69; 14.87 -> 27.73 | Helped |
| Train NeRF directly on linear raw images rather than tonemapped LDR images | NeRF · Synthetic Lego with clean inputs (infinite shutter speed) | sRGB PSNR: 33.16 -> 32.15 | Hurt |
| Train on raw images instead of LDR images | NeRF · Real denoising dataset (3 scenes) | Affine-aligned sRGB PSNR: 19.43 -> 23.53 | Helped |
| Pre-denoise training images with Unprocess (ablation) | RawNeRF · Real denoising dataset (3 scenes) | Affine-aligned sRGB PSNR: 23.53 -> 23.35 | Hurt |
| Pre-denoise training images with Unprocess (ablation) | RawNeRF · Real denoising dataset (3 scenes) | Affine-aligned sRGB SSIM: 0.536 -> 0.531 | Hurt |
| Pre-denoise training images with Unprocess (ablation) | RawNeRF · Real denoising dataset (3 scenes) | Affine-aligned sRGB LPIPS: 0.501 -> 0.507 | No clear effect |
| Use gradient-reweighted L2 rather than reweighted L1 or the camera-noise negative log-likelihood (ablation) | RawNeRF loss · RawNeRF training | Qualitative reconstruction quality | Helped |
| Use standard unweighted L2 or L1 instead of gradient-reweighted L2 (ablation) | RawNeRF loss · Very noisy scenes | Training stability | Hurt |
| Apply the unclipped sRGB gamma curve directly in the loss (ablation) | RawNeRF loss · Dark regions | Color artifacts | Hurt |
| Apply the log tone curve directly before L2 rather than reweighting by its gradient (ablation) | RawNeRF loss · RawNeRF training | Training stability | Hurt |
| Supervise all color channels using bilinearly demosaicked raw images instead of masking to the measured Bayer channel (ablation) | RawNeRF · High-noise image regions | Color artifacts | Hurt |
Closest papers by meaning
- NeRF On-the-go: Exploiting Uncertainty for Distractor-free NeRFs in the Wild (CVPR 2024 · similarity 0.92)
- Bright-NeRF: Brightening Neural Radiance Field with Color Restoration from Low-Light RAW Images (AAAI 2025 · similarity 0.92)
- LuSh-NeRF: Lighting up and Sharpening NeRFs for Low-light Scenes (NeurIPS 2024 · similarity 0.91)
- NeRF-VPT: Learning Novel View Representations with Neural Radiance Fields via View Prompt Tuning (AAAI 2024 · similarity 0.91)
- Pano-NeRF: Synthesizing High Dynamic Range Novel Views with Geometry from Sparse Low Dynamic Range Panoramic Images (AAAI 2024 · similarity 0.90)
- Enhancing Neural Radiance Fields with Adaptive Multi-Exposure Fusion: A Bilevel Optimization Approach for Novel View Synthesis (AAAI 2024 · similarity 0.90)
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.