Visual document search improves storage with generative embeddings
Generative Late-Interaction Embeddings For Visual Document Retrieval
Information Retrieval
Summary
Searching one thousand visual data vectors per page uses a lot of space and slows tools down. The authors discovered that these vectors cluster near a shape with only five to six dimensions and lie on a fixed-radius sphere. They created a method called GLIE that stores just a few summary vectors and can recreate the full detailed set when needed. This approach keeps most of the search accuracy but drastically reduces storage needs and speeds up retrieval.
What this means in practice
- •For document management teams: Lower storage and speed up visual document searches by storing a few generative vectors and reconstructing full details only for top matches.
- •For digital library operators: Store large visual document collections efficiently while maintaining high search accuracy using generative embeddings and on-demand reconstruction.
Authors
Mohamed Eltahir, Talal Aloushan, Rose Khairoalsendi, Jana Shata, Mohammed Alhassan, Leen Alrehaili, Tanveer Hussain, Naeemullah Khan
Abstract
Late-interaction retrieval is the state-of-the-art for visual document search, but it pays for its accuracy in storage. Existing compression methods retain a subset or local average of the N~1,000 vectors per page. Under aggressive storage budgets, however, these methods degrade sharply, and alternatives require retraining the encoder. Investigating this degradation across three encoders, we found two consistent properties: the vectors lie exactly on the unit sphere and concentrate near a manifold of intrinsic dimension five to six. This geometry yields two insights. First, standard k-means centroids fall inside the sphere, causing systematic underestimation of MaxSim scores. Normalizing them to the surface is a free correction worth up to +0.093 nDCG@5 over raw centroids. Second, because the page manifold has few degrees of freedom, the full set of vectors can be regenerated from only a few. To this end, we introduce Generative Late-Interaction Embeddings (GLIE): k << N vectors per page learned from the normalized centroids to serve as both a lightweight index and a basis for regenerating the page's full embedding set. At query time, search runs exclusively on these k vectors, and a decoder expands only the top candidates back to all N vectors for exact rescoring. At four vectors per page on ViDoRe v1, GLIE retains nearly 80% of the uncompressed system's nDCG@5, against 70% for the best prior post-hoc method. These results use a 415K-parameter network fitted in under three GPU-minutes on just a thousand training pages. At a matched training budget, fine-tuning the encoder does not reach even the training-free stage of GLIE, and the full system beats it at every budget. These patterns hold across a second encoder and ViDoRe v2. By reconstructing evidence on demand rather than sampling it, GLIE opens a new axis for storage-efficient retrieval, with the decoder as its main design surface.