RefGC-SR$^2$: Reference-guided Generated Content Super-Resolution and Refinement
Researchers have introduced a new task called reference-guided generated content super-resolution-refinement (RefGC-SR$^2$). This method aims to improve the quality of generated images by reusing the original high-resolution reference image during post-processing. The goal is to recover lost details, refine artifacts introduced by generative models, and upscale the output simultaneously. A new dataset and a frequency-aware diffusion transformer model have been developed to address this task, showing significant improvements over existing methods in terms of object identity and detail recovery. AI
IMPACT This research introduces a new method to improve the fidelity and detail of AI-generated images by leveraging high-resolution references.