Research map / Image generation / Fast diffusion sampling
Mitigating Noise Shift in Denoising Generative Models with Noise Awareness Guidance
Jincheng Zhong, Boyuan Jiang, Xin Tao, Pengfei Wan, Kun Gai, Mingsheng Long
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 NAG to converged DiT-XL/2 sampling without CFG | DiT-XL/2 · ImageNet 256x256 | FID: 9.62 -> 2.59 | Improved |
| add NAG to converged SiT-XL/2 sampling without CFG | SiT-XL/2 · ImageNet 256x256 | FID: 8.61 -> 2.26 | Improved |
| add NAG on top of CFG | SiT-XL/2 · ImageNet 256x256 | FID: 2.06 -> 1.72 | Improved |
| add NAG on top of CFG | DiT-XL/2 · ImageNet 256x256 | FID: 2.27 -> 2.14 | Improved |
| add NAG without CFG | DiT-XL/2 · ImageNet 256x256 | recall: 0.67 -> 0.60 | Hurt |
| add NAG to fine-tuned DiT-XL/2 without CFG | DiT-XL/2 · 7 fine-tuning datasets | average FID: 18.65 -> 13.72 | Improved |
| add NAG on top of CFG in fine-tuning | DiT-XL/2 · 7 fine-tuning datasets | average FID: 13.69 -> 10.31 | Improved |
| add NAG on top of Domain Guidance (DoG) | DiT-XL/2 · 7 fine-tuning datasets | average FID: 11.55 -> 10.05 | Improved |
| add NAG on top of DoG on Stanford Car | DiT-XL/2 · Stanford Car | FID: 4.38 -> 4.21 | No clear effect |
Closest papers by meaning
- Is Noise Conditioning Necessary for Denoising Generative Models? (ICML 2025 · similarity 0.92)
- NoiseCtrl: A Sampling-Algorithm-Agnostic Conditional Generation Method for Diffusion Models (CVPR 2025 · similarity 0.90)
- Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Denoising Task Difficulty-based Curriculum for Training Diffusion Models (ICLR 2025 · similarity 0.89)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.89)
- An Edit Friendly DDPM Noise Space: Inversion and Manipulations (CVPR 2024 · 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.
