Reproduce “Learning Dual-Level Deformable Implicit Representation for Real-World Scale Arbitrary 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
Zhiheng Li, Muheng Li, Jixuan Fan, Lei Chen, Yansong Tang, Jiwen Lu, Jie Zhou
Researchers at Tsinghua University developed the Dual-level Deformable Implicit Representation (DDIR) model for real-world scale arbitrary super-resolution. This work also introduces RealArbiSR, the first dataset for this task, enabling DDIR to produce high-fidelity high-resolution images from complex real-world low-resolution inputs at any desired magnification.
Key insights
- Real-world image degradations introduce dual-level discrepancies: global photometric variations (image-level deformation) and localized RGB mismatches due to spatially variant kernels (pixel-level deformation).
- Explicitly modeling these dual-level deformations within an implicit neural representation framework allows for accurate high-resolution reconstruction from real-world low-resolution images at arbitrary scales.
- A dedicated dataset with real-world low-resolution/high-resolution pairs at both integer and non-integer arbitrary scales is crucial for training and evaluating models capable of robust generalization in practical applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.