Research map / Image generation / Fast diffusion sampling
Laplacian Multi-scale Flow Matching for Generative Modeling
Zelin Zhao, Petr Molodyk, Haotian Xue, Yongxin 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 |
|---|---|---|---|
| use MoT shared model instead of separate per-scale models (ablation) | LapFlow · CelebA-HQ 256x256 | GFLOPs / FID: GFLOPs 38.9 -> 16.5; FID 3.60 -> 3.53 | Improved |
| add causal attention mask across scales (vs none) (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 3.91 -> 3.53 | Improved |
| use causal mask instead of self mask (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 5.19 -> 3.53 | Improved |
| set critical time T to 0.5 rather than 0.1 or 0.9 (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 5.12 (T=0.1), 4.92 (T=0.9) -> 3.53 (T=0.5) | Improved |
| use EQVAE instead of SDVAE (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 4.37 -> 3.53 | Improved |
| use EQVAE instead of SDVAE (ablation) | LFM baseline · CelebA-HQ 256x256 | FID: 5.26 -> 7.77 | Hurt |
| linear noise schedule instead of GVP (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 4.10 -> 3.53 | Improved |
| linear noise decay instead of cubic decay (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 4.25 -> 3.53 | Improved |
| two Laplacian scales instead of four (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 5.12 -> 3.53 | Improved |
| three scales instead of two at 256x256 (ablation) | LapFlow · CelebA-HQ 256x256 | FID: 3.53 -> 3.59 | No clear effect |
| three scales instead of two at 512x512 (ablation) | LapFlow · CelebA-HQ 512x512 | FID: 5.45 -> 4.04 | Improved |
| three scales instead of two at 1024x1024 (ablation) | LapFlow · CelebA-HQ 1024x1024 | FID: 6.62 -> 5.51 | Improved |
Closest papers by meaning
- Improving Progressive Generation with Decomposable Flow Matching (NeurIPS 2025 · similarity 0.89)
- Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation (NeurIPS 2026 · similarity 0.89)
- Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation (NeurIPS 2025 · similarity 0.88)
- Value Gradient Guidance for Flow Matching Alignment (NeurIPS 2025 · similarity 0.88)
- Align Your Flow: Scaling Continuous-Time Flow Map Distillation (NeurIPS 2025 · similarity 0.87)
- Bellman Optimal Stepsize Straightening of Flow-Matching Models (ICLR 2024 · 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.
