Research map / Image generation / Fast diffusion sampling
Frequency-Aware Flow Matching for High-Quality Image Generation
Sucheng Ren, Qihang Yu, Ju He, Xiaohui Shen, Liang-Chieh Chen
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 low-frequency component (ablation) | FreqFlow-B · ImageNet-256 ablation | FID: 3.86 -> 3.55 | Improved |
| add high-frequency component (ablation) | FreqFlow-B · ImageNet-256 ablation | FID: 3.86 -> 3.12 | Improved |
| add both low- and high-frequency components (ablation) | FreqFlow-B · ImageNet-256 ablation | FID: 3.86 -> 2.95 | Improved |
| add both low- and high-frequency components (ablation) | FreqFlow-B · ImageNet-256 ablation | IS: 200.4 -> 231.5 | Improved |
| merge by addition instead of cross attention (ablation) | FreqFlow-B · ImageNet-256 ablation | FID: 3.95 -> 2.95 | Improved |
| merge by addition instead of channel concatenation (ablation) | FreqFlow-B · ImageNet-256 ablation | FID: 3.46 -> 2.95 | Improved |
| add loss on frequency branch (ablation) | FreqFlow-B · ImageNet-256 ablation | FID: 4.67 -> 2.95 | Improved |
| add loss on frequency branch (ablation) | FreqFlow-B · ImageNet-256 ablation | IS: 198.4 -> 231.5 | Improved |
| use unified frequency branch instead of separate networks (ablation) | FreqFlow-B · ImageNet-256 ablation | FID: 3.44 -> 2.95 | Improved |
| vary loss weight alpha (ablation) | FreqFlow-B · ImageNet-256 | FID | No clear effect |
| FreqFlow vs SiT frequency error (high) | FreqFlow · frequency analysis | high-frequency error: 0.69 -> 0.48 | Improved |
| FreqFlow vs SiT frequency error (low) | FreqFlow · frequency analysis | low-frequency error: 0.08 -> 0.06 | 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.
- DiffiT — DiffiT: Diffusion Vision Transformers for Image Generation (ECCV 2024) · better on 1 of 1 shared result
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
- Balancing Frequencies and Pixels in Flow Matching (NeurIPS 2026 · similarity 0.91)
- Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation (NeurIPS 2025 · similarity 0.89)
- Improving Progressive Generation with Decomposable Flow Matching (NeurIPS 2025 · similarity 0.88)
- Structured Coupling for Flow Matching (NeurIPS 2026 · similarity 0.88)
- Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation (NeurIPS 2026 · similarity 0.88)
- Gaussian Mixture Flow Matching Models (ICML 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.
