One email each morning — yesterday's AI, sortedGet it in your inbox

METAL LAB

PixRestore: Unified Image Restoration via Pixel Diffusion Transformer

arXiv:2608.167932026-08-16

Unified image restoration (UIR) aims to recover high-quality (HQ) content from low-quality (LQ) images with different degradations using a single model. Most recent methods adapt large pretrained text-to-image (T2I) latent diffusion models for their strong capacity and generative priors. However, the variational autoencoder (VAE) in latent T2I models may discard restoration-sensitive details, while the open-ended synthesis prior can introduce content-inconsistent artifacts. We present PixRestore

Authors · Lingchen Sun

Read on arXiv

Latest papers

All papers →

Latest from METAL LAB