Research map / Image generation / Fast diffusion sampling
Flow Straight to Reality: Perceptually Consistent Flow Matching for Efficient Image Restoration
Sangwoo (Jason) Jo, Donggeun Ko, Jayeon Kang, Youngsang Kwak, Jaehwa Kwak, Sungjoon Choi
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 |
|---|---|---|---|
| replace ELIR with PCFlow (LCPL + conflict-free alignment) | ELIR · BFR CelebA-Test | FID: 44.64 -> 35.89 | Improved |
| replace ELIR with PCFlow across other tasks | ELIR · SR/denoising/inpainting/colorization | FID: e.g. 51.72 -> 45.21 (colorization) | Improved |
| use PCFlow instead of PMRF (K=25) on other tasks | PMRF · SR/denoising/inpainting/colorization | FID/PSNR: FID 44.64 (PMRF) vs 45.50 on SR | Hurt |
| use PCFlow instead of PMRF on BFR distortion | PMRF · BFR CelebA-Test | PSNR: 25.85 -> 24.44 | Hurt |
| use conditional flow from noise instead of unconditional flow (ablation) | PCFlow · Colorization | FID: 56.27 -> 47.21 | Improved |
| use conditional flow instead of unconditional on super-resolution (ablation) | PCFlow · Super-resolution | FID: 63.69 -> 46.42 | Improved |
| fine-tune encoder jointly with vector field (ablation) | PCFlow · Colorization | FID: 47.21 -> 45.97 | Improved |
| add latent consistency perceptual loss (ablation) | PCFlow · Colorization | FID: 45.97 -> 45.78 | Improved |
| add conflict-free gradient alignment (ablation) | PCFlow · Colorization | FID: 45.78 -> 45.21 | Improved |
| add 250-epoch preheating before perceptual loss (ablation) | PCFlow · Super-resolution | FID: 46.26 -> 45.50 | Improved |
| add perceptual loss from epoch 0 without preheating (ablation) | PCFlow · Super-resolution | FID: 46.10 -> 46.40 | No clear effect |
| use linear warmup lambda schedule instead of constant (ablation) | PCFlow · Super-resolution | FID: 46.18 -> 45.50 | Improved |
Closest papers by meaning
- Posterior-Mean Rectified Flow: Towards Minimum MSE Photo-Realistic Image Restoration (ICLR 2025 · similarity 0.89)
- Reversing Flow for Image Restoration (CVPR 2025 · similarity 0.89)
- GenRec: Knowing Where to Reconstruct and Where to Generate (NeurIPS 2026 · similarity 0.87)
- Improving Progressive Generation with Decomposable Flow Matching (NeurIPS 2025 · similarity 0.87)
- Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation (NeurIPS 2026 · similarity 0.87)
- Balancing Frequencies and Pixels in Flow Matching (NeurIPS 2026 · similarity 0.86)
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.
