Reproduce “Low-light Image Restoration with Short- and Long-exposure Raw Pairs” 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
Meng Chang, Huajun Feng, Zhihai Xu, Qi Li
Researchers at Zhejiang University developed a method to restore low-light images on mobile devices by fusing raw short- and long-exposure image pairs using a simulated imaging pipeline for data generation and a specialized neural network. Their approach yielded a PSNR of 30.70 dB on synthetic data and consistently produced sharp, color-accurate, and artifact-free images in real-world tests, outperforming existing methods.
Key insights
- Processing directly in the raw sensor data domain is critical for superior low-light image restoration, as it preserves more information and simplifies noise distribution compared to sRGB domain processing.
- Accurately simulating the entire camera imaging pipeline, from light capture to raw data generation, is essential for creating high-fidelity synthetic training data that enables models to generalize well to real-world conditions.
- A unified network architecture that jointly tackles multiple degradations (noise, blur, color distortion, misalignment) through specialized modules like alignment and deghosting is more effective than sequential or isolated approaches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.