Zhuchenyang Liu

The work frames inversion of a multi-vector visual document index as conditional document image generation, identifying the encoder, inferring page shape and restoring shuffled order from the stored vectors alone under zero side information, and redrawing the page with a conditional flow-matching inverter; on ViDoRe v3, pages inverted from the raw index recover 47% of words and 45% of sensitive tokens and rank their source page first 98.4% of the time.

The work frames inversion of a multi-vector visual document index as conditional document image generation, identifying the encoder, inferring page shape and restoring shuffled order from the stored vectors alone under zero side information, and redrawing the page with a conditional flow-matching inverter; on ViDoRe v3, pages inverted from the raw index recover 47% of words and 45% of sensitive tokens and rank their source page first 98.4% of the time.

The work introduces ColNanoVDR, which uses OTW, an entropic optimal-transport objective with learned token weights, to distill the query encoder of multi-vector visual document retrieval teachers into 149M text-only students without encoding or reading any page during training, retaining about 95% of the teachers' NDCG@5 on ViDoRe v1-v3, encoding queries up to 26x faster on a single CPU thread, and matching score distillation under identical training while reading 12.6x less cached teacher data.

The work introduces ColNanoVDR, which uses OTW, an entropic optimal-transport objective with learned token weights, to distill the query encoder of multi-vector visual document retrieval teachers into 149M text-only students without encoding or reading any page during training, retaining about 95% of the teachers' NDCG@5 on ViDoRe v1-v3, encoding queries up to 26x faster on a single CPU thread, and matching score distillation under identical training while reading 12.6x less cached teacher data.