Reproduce “Linear Recurrent Unit with Semantic Modulation for Image Super-Resolution” 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
Mingyu Choi, Woo Kyoung Han, Sunghoon Im, Kyong Hwan Jin
A new architecture for image super-resolution, Linear Recurrent Unit with Semantic Modulation (LSM), effectively integrates Linear Recurrent Units (LRUs) with pixel-wise semantic modulation. This approach achieves competitive reconstruction quality while significantly reducing computational complexity and memory usage compared to prior methods.
Key insights
- Careful initialization of LRU eigenvalues, particularly a broad phase range (θ_max=2π), is crucial for effective long-range modeling and fine texture reconstruction in image SR.
- Pixel-wise, input-dependent semantic modulation successfully transforms the static linear recurrence of LRUs into an adaptive mechanism capable of handling complex 2D visual features prevalent in image super-resolution.
- A multi-functional Semantic Modulating Unit (SMU) combining LRU modulation, semantic categorization, and global feature enhancement synergistically improves both long-range contextual understanding and local detail reconstruction within an efficient single-scan paradigm.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.