Research map / Probabilistic and scientific modeling / Variational and generative sampling
Wavelet Flow: Fast Training of High Resolution Normalizing Flows
Jason J. Yu, Konstantinos G. Derpanis, Marcus A Brubaker
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 |
|---|---|---|---|
| wavelet-based multi-scale conditional flow instead of Glow | Wavelet Flow · ImageNet 64x64 | bits per dimension: 3.81 -> 3.78 | Improved |
| wavelet-based multi-scale conditional flow instead of Glow | Wavelet Flow · LSUN bedroom 64x64 | bits per dimension: 2.38 -> 2.41 | Hurt |
| wavelet-based flow instead of Glow (training cost) | Wavelet Flow · CelebA-HQ 256x256 | seconds per image: 1.79 -> 0.0147 | Improved |
| wavelet-based flow instead of Glow (total training) | Wavelet Flow · LSUN bedroom 64x64 | training GPU hours: 3,700 -> 572 | Improved |
| wavelet-based flow instead of Glow (sample quality) | Wavelet Flow · LSUN bedroom 64x64 | FID: 60.03 -> 121.20 | Hurt |
| use affine instead of additive coupling (ablation) | Wavelet Flow · CelebA-HQ 256x256 5-bit | bits per dimension: 1.12 -> 0.94 | Improved |
| use affine instead of additive coupling for annealed samples (ablation) | Wavelet Flow · CelebA-HQ | image quality | No clear effect |
| truncate 64x64 model to evaluate at 32x32 (ablation) | Wavelet Flow · ImageNet 32x32 | bits per dimension: 4.08 -> 4.16 | Hurt |
| use 5-bit instead of 8-bit data to improve samples | Wavelet Flow · CelebA-HQ | sample quality | No clear effect |
| add multi-scale MCMC annealing at T=0.97 (ablation) | Wavelet Flow · CelebA-HQ | sample quality | Improved |
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.
- Glow — Glow: Generative Flow with Invertible 1x1 Convolutions (NeurIPS 2018) · better on 6 of 9 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
- STARFlow: Scaling Latent Normalizing Flows for High-resolution Image Synthesis (NeurIPS 2025 · similarity 0.87)
- Building Normalizing Flows with Stochastic Interpolants (ICLR 2023 · similarity 0.87)
- Normalizing Flows: An Introduction and Review of Current Methods (TPAMI 2020 · similarity 0.87)
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.87)
- Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (ICML 2024 · similarity 0.85)
- Analysis of Learning a Flow-based Generative Model from Limited Sample Complexity (ICLR 2024 · similarity 0.84)
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.
