MAGIC：基于最优传输的边际引导压缩用于高效视觉文档检索
原标题：MAGIC: Marginal-Guided Compression with Optimal Transport for Efficient Visual Document Retrieval
AI 摘要
该论文提出 MAGIC，一种无需训练的事后压缩方法，用于提升视觉文档检索（VDR）效率。针对 ColPali 等多向量检索系统中 patch 级向量带来的存储与 MaxSim 计算开销，MAGIC 通过估计检索需求并构建双边际熵最优传输问题，在压缩时优先保留高使用率 patch 并均衡保留 facet 的使用。在 ViDoRe 基准上，MAGIC 在多种保留比例和检索骨干下优于现有事后压缩器，尤其在激进压缩场景提升明显，代码已开源。
正文节选
MAGIC: Marginal-Guided Compression with Optimal Transport for Efficient Visual Document Retrieval Abstract Recent visual document retrieval (VDR) systems such as ColPali use multi-vector page embeddings, in which patch-level vectors enable fine-grained evidence matching but incur substantial index storage and MaxSim scoring overhead. Post-hoc merging offers a practical route to efficient VDR by reducing this cost without retraining the retriever, but its uniform reconstruction objectives are poo