Robust-U1: MLLMs Recover Corrupted Images First
Robust-U1 trains an MLLM to reconstruct corrupted visual content, reaching 0.7398 overall on R-Bench versus 0.5770 for BAGEL and 0.5017 for Robust-R1.
Institution
Nanjing University of Science and Technology
A Chinese research university in Nanjing with work across engineering, computer vision, and applied machine learning.