Visual Text Compression as Measure Transport

Lv Tang, Tianyi Zheng, Yang Liu +2

Visual text compression (VTC) promises efficient long-context processing by rendering text into an image and re-encoding it with a vision-language model, often producing $3$--$20\t…