Reproduce “Learning Latent Transmission and Glare Maps for Lens Veiling Glare Removal” 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
Xiaolong Qian, Qi Jiang, Lei Sun, Zongxi Yu, Kailun Yang, Peixuan Wu, Jiacheng Zhou, Yao Gao and 3 more
A framework named DeVeiler was developed by researchers at Zhejiang University and collaborators to computationally remove co-existing optical aberrations and veiling glare from images captured by simplified optical systems. It leverages a physics-informed generative model, VeilGen, to synthesize realistic paired training data, enabling a restoration network to recover color fidelity and fine details and surpass prior methods on both synthetic and real-world degraded images.
Key insights
- Complex optical degradations like veiling glare, which are difficult to simulate, can be effectively learned and synthesized by physics-informed generative models, addressing the critical data scarcity problem.
- Distilling the behavior of a multi-step generative model into a lightweight degradation network provides an efficient and effective means to supervise a restoration network for compound degradations.
- Enforcing reversibility between the learned forward degradation and the restoration process through symmetrically designed modules (VGIM/VGCM) enables the restoration network to learn physically interpretable latent maps for targeted glare removal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.