Reproduce “DynFaceRestore: Balancing Fidelity and Quality in Diffusion-Guided Blind Face Restoration with Dynamic Blur-Level Mapping and Guidance” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Huu-Phu Do, Yu-Wei Chen, Yi-Cheng Liao, Chi-Wei Hsiao, Han-Yang Wang, Wei-Chen Chiu, Ching-Chun Huang
Researchers from National Yang Ming Chiao Tung University and MediaTek Inc. developed DynFaceRestore, a diffusion-guided framework for blind face restoration that dynamically maps blur levels, selects starting diffusion timesteps, and applies region-aware guidance. This approach achieves enhanced perceptual quality and preserves facial identity across diverse degradation severities on both synthetic and real-world datasets.
Key insights
- Re-framing blind degradation as a Gaussian deblurring problem, coupled with dynamic kernel refinement, provides a robust method for mitigating kernel mismatch in real-world scenarios.
- Dynamically selecting the diffusion starting timestep based on input blur levels is crucial for preventing under- or over-diffusion and optimizing the generative process.
- Applying region-specific guidance allows the diffusion model to intelligently enhance details in high-frequency areas while maintaining structural fidelity in low-frequency regions of the face.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.