MarginMerge Compression: Preserves 97% Quality for Visual Document Retrievers with 90% Fewer Vectors
_reachsumit · x · 2026-08-05
Multi-vector visual document retrievers like ColPali produce large indexes and costly scoring due to storing fine-grained patch embeddings.
This paper introduces MarginMerge, a compression method that selects coverage-aware anchors, clusters document patches, and synthesizes representative vectors. Experiments show that while reducing stored document vectors by 90% to 95%, this method preserves 97% to 99% of the average nDCG@5 retrieval quality compared to the uncompressed index.
More from Research
- Agent-formalized preprints found to be flaw-free in small sample test — Sauers_ · 2026-08-27
- Britannica Illustrated Pages Dataset Hits Hugging Face — biglam · 2026-08-27
- VBVR-Pro: A Scalable Suite for Native Visual Reasoning — Junxiang Xu · 2026-08-27
- Code World Model: Coding Agent as World Brain — Yiwen Chen · 2026-08-27
- PISA: A Tool for Visualizing Cis-Regulatory Rules in Genomic Data — anshulkundaje · 2026-08-27
- CHI Papers Lack Prediction-Powered Inference for LLM Evaluation — IanArawjo · 2026-08-27