Reproduce “Fidelity- and Perception-Aware Local Implicit Attention for Arbitrary-Scale 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
Yu-Syuan Xu, Hao-Lun Sun, Hao-Wei Chen, Hsien-Kai Kuo, Chun-Yi Lee
Fidelity- and Perception-Aware Local Implicit Attention (FPLIA) is introduced for arbitrary-scale image super-resolution, adeptly balancing reconstruction fidelity and perceptual quality by adaptively integrating features from pixel-regression and diffusion models. This framework achieves Pareto-optimal performance, outperforming existing methods in visual realism without sacrificing structural accuracy.
Key insights
- Fidelity- and perceptual-oriented features provide distinct yet complementary information; their productive interaction can be asymmetric, where fidelity guides perceptual features towards faithfulness, and perceptual features enrich smooth fidelity representations with high-frequency diversity.
- Discrete, per-pixel selection of feature candidates is superior to soft blending or global weighting, allowing content- and scale-aware adaptation that prevents feature degradation from averaging across different statistical manifolds.
- Multi-branch supervision during training, where each candidate feature and the final selected output are independently supervised, ensures that all generated candidates remain informative and contribute to a balanced fidelity-perception outcome.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.