Search papers, labs, and topics across Lattice.
This paper introduces DocPC, a novel framework for document-level visual retrieval that enhances efficiency by utilizing Representative Page Composition to create a single grid image from selected representative pages. By addressing the limitations of existing page-centric methods, DocPC reduces indexed images and storage requirements by 10.1x and improves indexing speed by approximately 7.7x. The framework's effectiveness is validated through the DocViRe benchmark, achieving an NDCG@5 score of 44.09, significantly surpassing the best page-level baseline.
DocPC slashes indexing time by 7.7x while boosting retrieval accuracy, redefining how we approach document-level visual search.
Visual document retrieval has advanced by encoding page screenshots with vision-language models, bypassing OCR pipelines. However, existing methods remain page-centric, misaligned with real-world scenarios requiring complete document retrieval. A naive page-then-document aggregation suffers from linear indexing cost and degraded retrieval when relevance spans multiple pages. We propose DocPC, a document-level visual retrieval framework based on Representative Page Composition: selecting representative pages and composing them into a single grid image for document-level indexing, reducing indexed images, vectors, and storage by 10.1x and end-to-end indexing time by roughly 7.7x. To handle multi-positive supervision prevalent at the document level, we combine multi-positive contrastive learning with sparsely scheduled listwise optimization. We also introduce DocViRe, a benchmark with multi-positive relevance annotations. DocPC-ColQwen achieves NDCG@5 of 44.09 on DocViRe, outperforming the strongest page-level baseline at 38.91 while reducing storage by 10.1x. Code is available at https://anonymous.4open.science/r/DocPC-Document-Level-Visual-Retrieval-via-Representative-Page-Composition-1D52. Data is available at https://huggingface.co/datasets/anonymous-7219/docpc.
