Research map / Image generation / Image restoration
YeTI: You Only Need Two Noisy Images for Real-World sRGB Noise Generation
Jaekyun Ko, Byung Wan Lim, Dongjin Kim, Soomin Lee, Tae Hyun Kim
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 disentangled structure+noise conditioning in C-DiT (vs single encoder) (ablation) | YeTI C-DiT · MAI2021 | KLD: 0.1584 -> 0.0793 | Improved |
| add L_Norm on noise latent (ablation) | YeTI RAE · SIDD val metadata classification | Top-1 accuracy (%): 47.20 -> 65.22 | Improved |
| add contrastive projection head (ablation) | YeTI RAE · SIDD val metadata classification | Top-1 accuracy (%): 51.20 -> 65.22 | Improved |
| add contrastive loss L_Cont (ablation) | YeTI RAE · SIDD val structure purity | PSNR: 15.49 -> 23.06 | Improved |
| add projection head (ablation) | YeTI RAE · SIDD val structure purity | PSNR: 17.21 -> 23.06 | Improved |
| add total variation loss L_TV (ablation) | YeTI RAE · SIDD val structure purity | PSNR: 21.49 -> 23.06 | Improved |
| use subtractive-additive injection instead of element-wise addition (ablation) | YeTI RAE · SIDD val structure purity | PSNR: 12.70 -> 23.06 | Improved |
| use subtractive-additive injection instead of concatenation (ablation) | YeTI RAE · SIDD val structure purity | PSNR: 14.42 -> 23.06 | Improved |
| mix 50% YeTI synthetic with real noisy images for training | AP-BSN · SIDD val denoising | PSNR: 36.86 -> 37.09 | Improved |
| mix 50% YeTI synthetic with real noisy images for training | AP-BSN · DND denoising | PSNR: 38.37 -> 38.55 | Improved |
| raise lambda_TV and lambda_Norm from 1e-6 to 1e-5 (ablation) | YeTI RAE · SIDD val metadata classification | Top-1 accuracy (%): 65.22 -> 42.71 | Hurt |
| shift burst frames by N=4 pixels during training (ablation) | YeTI · SIDD val | KLD: 0.0256 -> 0.0416 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SeNM-VAE (0.01%) — SeNM-VAE: Semi-Supervised Noise Modeling with Hierarchical Variational Autoencoder (CVPR 2024) · better on 3 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
- Diffusion-Based sRGB Real Noise Generation via Prompt-Driven Noise Representation Learning (CVPR 2026 · similarity 0.91)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.90)
- sRGB Real Noise Synthesizing With Neighboring Correlation-Aware Noise Model (CVPR 2023 · similarity 0.90)
- Modeling sRGB Camera Noise With Normalizing Flows (CVPR 2022 · similarity 0.90)
- Rethinking Noise Synthesis and Modeling in Raw Denoising (ICCV 2021 · similarity 0.89)
- Realistic Noise Synthesis with Diffusion Models (AAAI 2025 · similarity 0.87)
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.
