Reproduce “BurstGP: Enhancing Raw Burst Image Super Resolution with Generative Priors” 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
Dong Huo, Tristan Aumentado-Armstrong, Samrudhdhi B. Rangrej, Maitreya Suin, Angela Ning Ye, Zhiming Hu, Amanpreet Walia, Amirhossein Kazerouni and 3 more
A novel diffusion-based framework, BurstGP, enhances raw burst image super-resolution by integrating pre-trained video diffusion models and robustly bridging the lRGB-sRGB domain gap. The method consistently improves perceptual quality on synthetic and real-world datasets while maintaining fidelity, for instance, boosting TOPIQ by 25% on RealBSR-RAW.
Key insights
- Leveraging pre-trained *video* diffusion models as multi-frame aware refiners significantly boosts perceptual quality in raw burst image super-resolution.
- Dynamic degradation conditioning is effective for adaptively controlling the generative prior's strength, enabling a balanced perception-distortion trade-off.
- A robust inverse ISP is essential for stable and physically consistent conversion of sRGB diffusion outputs back to the linear RGB domain for raw image processing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.