Research map / Image generation / Fast diffusion sampling
Spectrally-Guided Diffusion Noise Schedules
Carlos Esteves, Ameesh Makadia
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 |
|---|---|---|---|
| use per-instance spectrally-guided mixed noise schedule instead of SiD2 schedule (with half the steps) | SiD2 small · ImageNet 256x256 | FID: 1.68 -> 1.42 | Improved |
| use per-instance spectrally-guided schedule | SiD2 flop heavy · ImageNet 512x512 | FID: 1.53 -> 1.45 | Improved |
| use per-instance spectrally-guided schedule | SiD2 small · MS-COCO 512x512 zero-shot | FID: 7.56 -> 7.08 | Improved |
| use per-instance schedule instead of a fixed median schedule (ablation) | Ours · ImageNet 256x256 | FID: 1.61 -> 1.42 | Improved |
| use spectral min/max-bounded mixed schedule instead of cosine between per-instance min and max (ablation) | Ours · ImageNet 256x256 | FID: 1.98 -> 1.42 | Improved |
| average frequency-focused and power-focused schedules (mixed) instead of frequency-focused only (ablation) | Ours · ImageNet 256x256 | FID: 5.16 -> 1.42 | Improved |
| average frequency-focused and power-focused schedules (mixed) instead of power-focused only (ablation) | Ours · ImageNet 256x256 | FID: 5.18 -> 1.42 | Improved |
| add min/max logSNR FiLM conditioning (ablation) | Ours · ImageNet 256x256 | FID: 1.58 -> 1.42 | Improved |
| define guidance interval in t instead of logSNR (ablation) | Ours · ImageNet 256x256 | FID: 1.58 -> 1.42 | Improved |
| use ground-truth spectrum instead of trained RAPSD sampler (ablation) | Ours · ImageNet 256x256 | FID: 1.42 -> 1.43 | No clear effect |
| use ground-truth spectrum instead of trained RAPSD sampler (ablation) | Ours · MS-COCO 512x512 zero-shot | FID: 7.08 -> 6.55 | Improved |
| increase number of denoising steps at high counts (ablation) | Ours · ImageNet | FID: 7.08 -> 7.83 (MS-COCO, 96 -> 256 NFE) | Hurt |
Closest papers by meaning
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · similarity 0.92)
- Training-Free Adaptive Diffusion with Bounded Difference Approximation Strategy (NeurIPS 2024 · similarity 0.89)
- NoiseCtrl: A Sampling-Algorithm-Agnostic Conditional Generation Method for Diffusion Models (CVPR 2025 · similarity 0.89)
- Schedule On the Fly: Diffusion Time Prediction for Faster and Better Image Generation (CVPR 2025 · similarity 0.89)
- How I Warped Your Noise: a Temporally-Correlated Noise Prior for Diffusion Models (ICLR 2024 · similarity 0.89)
- One More Step: A Versatile Plug-and-Play Module for Rectifying Diffusion Schedule Flaws and Enhancing Low-Frequency Controls (CVPR 2024 · similarity 0.89)
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.
