Generalized and Efficient 2D Gaussian Splatting for Arbitrary-scale Super-Resolution

Du Chen, Liyi Chen, Zhengqiang Zhang, Lei Zhang · 2025

Read Generalized and Efficient 2D Gaussian Splatting for Arbitrary-scale Super-Resolution by Du Chen, Liyi Chen, Zhengqiang Zhang, Lei Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-06838.txt.

Description

Du Chen, Liyi Chen, Zhengqiang Zhang, and Lei Zhang published this paper in 2025. It brings Gaussian splatting, a technique that conquered 3D graphics, to image super-resolution at any scale. The task is to enlarge a low-resolution image by a factor the system has never seen, without the blur of fixed-ratio upscalers.

Instead of a neural field queried pixel by pixel, their model predicts a set of two-dimensional Gaussians fitted to the input image. Each is an oriented blob that can stretch along textures and edges. A custom GPU rasterizer samples the blobs at whatever output size is requested, giving smooth, scale-aware rendering. Trained end to end, the network generalizes across images and unseen magnifications. Tests showed strong quality against leading arbitrary-scale methods, with splatting's known speed advantages. Code and models are public.

Catalog record

CREATORDu Chen, Liyi Chen, Zhengqiang Zhang, Lei Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE95 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-06838

Browse the full library · About the Tower of Babel