Efficient Vision-Language Models by Summarizing Visual Tokens into Compact Registers

VLM
Re-assign community