Research map / Image generation / Fast diffusion sampling
Exposure Bias Can Alleviate Itself via Directional and Frequency Rectification in Flow Matching
Guanbo Huang, Jingjia Mao, Fanding Huang, fengkai liu, Xiangyang Luo, Yaoyuan Liang, Jiasheng Lu, xiaoe Wang, Pei Liu, Ruiliu Fu, Ruqi Huang, Shao-Lun Huang
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 DEFAR (ADR + FC) to SiT | SiT-B/4 · ImageNet-256 conditional (no CFG) | FID: 61.64 -> 56.39 | Improved |
| add Frequency Compensation to ADR (ablation) | DEFAR SiT-B/4 · ImageNet-256 conditional (no CFG) | FID: 59.77 -> 56.39 | Improved |
| add Anti-Drift Rectification to FC (ablation) | DEFAR SiT-B/4 · ImageNet-256 conditional (no CFG) | FID: 57.17 -> 56.39 | Improved |
| replace static-target SDSS with ADR | SiT-B/4 · ImageNet-256 conditional (no CFG) | FID: 61.25 -> 59.77 | Improved |
| add DEFAR to SiT with CFG | SiT-B/4 · ImageNet-256 conditional (with CFG) | FID: 31.64 -> 28.34 | Improved |
| add DEFAR to unconditional SiT | SiT-B/4 · CIFAR-10 unconditional | FID: 17.41 -> 14.82 | Improved |
| add DEFAR to unconditional SiT | SiT-B/4 · CelebA-64 unconditional | FID: 7.04 -> 4.53 | Improved |
| scale SiT-XL/2 with DEFAR | SiT-XL/2 · ImageNet-256 conditional (no CFG) | FID: 25.85 -> 18.45 | Improved |
| add DEFAR to discriminator guidance | SiT-G++ · ImageNet-256 conditional (no CFG) | FID: 60.92 -> 59.03 | Improved |
| add DEFAR to OT-CFM | OT-CFM · CIFAR-10 unconditional | Avg. FID: 4.095 -> 3.960 | Improved |
| add DEFAR to REPA-XL/2 | REPA-XL/2 · ImageNet-512 | FID: 3.99 -> 3.46 | Improved |
| normalize anti-drift target to unit length (ablation) | DEFAR · CIFAR-10 conditional | FID: 317.31 -> 11.22 | Improved |
Closest papers by meaning
- Simple ReFlow: Improved Techniques for Fast Flow Models (ICLR 2025 · similarity 0.87)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · similarity 0.87)
- PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation (NeurIPS 2026 · similarity 0.87)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.87)
- Elucidating Flow Matching ODE Dynamics via Data Geometry and Denoisers (ICML 2025 · similarity 0.87)
- Rectified Diffusion: Straightness Is Not Your Need in Rectified Flow (ICLR 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.
