Research map / Image generation / Fast diffusion sampling
Learning multi-scale local conditional probability models of images
Read the paperCheck 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 |
|---|---|---|---|
| Use multi-scale wavelet conditional denoising instead of a pixel-domain local Markov CNN | Face image denoising · CelebA, 160×160 | PSNR | Helped |
| Reduce conditional receptive fields to 9×9 (ablation) | Multi-scale wavelet denoiser · CelebA, 160×160 | PSNR | No clear effect |
| Reduce conditional receptive fields to 5×5 (ablation) | Multi-scale wavelet denoiser · CelebA, 160×160 | PSNR | Hurt |
| Reduce receptive fields in the pixel-domain CNN (ablation) | Conventional CNN denoiser · CelebA, 160×160 | PSNR | Hurt |
| Use multi-scale wavelet conditional sampling instead of pixel-domain local Markov sampling | Face image super-resolution · CelebA HQ, 320×320 | visual coherence | Helped |
| Use local pixel-domain CNN receptive fields smaller than the image (ablation) | Face image synthesis · CelebA HQ, 320×320 | global face structure | Hurt |
| Add a global-receptive-field low-pass CNN before local wavelet conditional stages | Multi-scale wavelet synthesis · CelebA HQ, 320×320 | global face structure | Helped |
Closest papers by meaning
- Locality in Image Diffusion Models Emerges from Data Statistics (NeurIPS 2025 · similarity 0.88)
- Generalization in diffusion models arises from geometry-adaptive harmonic representations (ICLR 2024 · similarity 0.86)
- Diffusion Prior Interpolation for Flexibility Real-World Face Super-Resolution (AAAI 2025 · similarity 0.86)
- Unsupervised Region-Based Image Editing of Denoising Diffusion Models (AAAI 2025 · similarity 0.86)
- Learning normalized image densities via dual score matching (NeurIPS 2025 · similarity 0.85)
- Likelihood Training of Cascaded Diffusion Models via Hierarchical Volume-preserving Maps (ICLR 2024 · similarity 0.85)
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.