Research map / Image generation / Autoregressive and latent generation
Geometry-Aware Image Flow Matching
Junho Lee, Kwanseok Kim, Joonseok Lee
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 |
|---|---|---|---|
| project targets onto hypersphere | I-CFM · CIFAR-10 | gFID: 4.29 -> 4.10 | Improved |
| project targets onto hypersphere | I-CFM · ImageNet-256 | gFID: 5.29 -> 5.02 | Improved |
| replace Euclidean cost with angular distance | OT-CFM · CIFAR-10 | gFID: 4.30 -> 4.11 | Improved |
| replace Euclidean cost with angular distance | OT-CFM · ImageNet-256 | gFID: 5.22 -> 5.15 | Improved |
| combine projection with SOT-CFM | SOT-CFM · ImageNet-256 | gFID: 5.15 -> 5.10 | Improved |
| move flow fully onto sphere (SFM) | I-CFM · ImageNet-256 | gFID: 5.29 -> 4.62 | Improved |
| move flow fully onto sphere (SFM) | I-CFM · CIFAR-10 | gFID: 4.29 -> 3.79 | Improved |
| increase projection radius from 30 to 120 (ablation) | SFM · ImageNet-256 | gFID: 4.95 -> 4.62 | Improved |
| increase projection radius from 120 to 130 (ablation) | SFM · ImageNet-256 | gFID: 4.62 -> 4.65 | Hurt |
| add ResNet50 norm prediction (ablation) | SFM · ImageNet-256 | gFID: 4.84 -> 4.81 | No clear effect |
| add ResNet50 norm prediction (ablation) | I-CFM · ImageNet-256 | gFID: 5.02 -> 4.98 | No clear effect |
| project data to mean norm (ablation) | SD3-VAE latent · ImageNet-256 | rFID: 0.21 -> 0.40 | Hurt |
Closest papers by meaning
- Categorical Flow Matching on Statistical Manifolds (NeurIPS 2024 · similarity 0.90)
- Beckmann Transport Models: From Autonomous Flows to One-Step Maps (NeurIPS 2026 · similarity 0.89)
- Flow Matching on General Geometries (ICLR 2024 · similarity 0.89)
- Metric Flow Matching for Smooth Interpolations on the Data Manifold (NeurIPS 2024 · similarity 0.88)
- Direct Product Flow Matching: Decoupling Radial and Angular Dynamics for Few-Shot Adaptation (NeurIPS 2026 · similarity 0.87)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · 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.
