Fetching the paper…

ViCToR: Improving Visual Comprehension via Token Reconstruction for Pretraining LMMs · Around