Training ESRGAN with multi-scale attention U-Net discriminator

Quan Chen, Hao Li, Gehao Lu · 2024

Read Training ESRGAN with multi-scale attention U-Net discriminator by Quan Chen, Hao Li, Gehao Lu (2024) 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/pmc-11585613.txt.

Description

Quan Chen and co-authors propose MSA-ESRGAN in 2024, a model that enlarges images while keeping them looking real. Super-resolution systems can sharpen a picture yet blur texture or mix subject and background. Their change sits in the part of the network that judges whether an image looks real or made.

That judging part uses a U-Net with attention at several scales, helping it tell subject from background more closely. Trained in the same setup as a known rival for fair comparison, the model was tested on five public image sets. Scores for image quality, signal strength, and structural likeness were higher than older methods, and human judges also rated the results better for texture and realism. A removal test showed the new judging part played a key role.

Catalog record

CREATORQuan Chen, Hao Li, Gehao Lu
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE49 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC11585613

Browse the full library · About the Tower of Babel