1 paper
Zhiyuan Zhao, Linke Ouyang, Bin Wang +5
Despite the great advance of Multimodal Large Language Models (MLLMs) in both instruction dataset building and benchmarking, the independence of training and evaluation makes curre…