New GB-LSR method offers faster, more efficient image reconstruction

Researchers have introduced GB-LSR, a novel local spectral image representation designed for efficient continuous image reconstruction and super-resolution. This method partitions images into patches, each using Fourier basis coefficients derived from shared convolutional features. A key innovation is the use of a single, globally shared bandwidth parameter, which simplifies the process and allows for reconstruction at any continuous coordinate with a fixed computational cost, independent of image size. Experiments show GB-LSR outperforms existing methods in native reconstruction benchmarks by significant margins in PSNR and LPIPS while operating at a fraction of the inference cost. AI

IMPACT This new representation could lead to more efficient and faster image processing in applications requiring high-fidelity reconstruction and super-resolution.

RANK_REASON The cluster describes a new research paper detailing a novel method for image reconstruction and super-resolution.

Read on arXiv cs.LG →

• GB-LSR
• Kodak
• LIIF-RDN
• LTE
• LTE-SwinIR
• Urban100
• WIRE
• Fourier
• Global-Bandwidth Local Spectral Representation

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →