Metraipolis › archive › dc_7d6f1316cfac
AdaMerge: Tuning-Free Patch Compression for Multi-Vector Visual Document Retrieval — arXiv:2609.22562
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.22562
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-09T21:58:51.776Z
- SHA-256 of the archived text
c0f2f27b2ea0abeb55a94ebab0fe248c6172da3d833e2685bf6c32697fd94c84- Stable URL of this page
https://metraipolis.com/documento/dc_7d6f1316cfac
How to cite
AdaMerge: Tuning-Free Patch Compression for Multi-Vector Visual Document Retrieval — arXiv:2609.22562 — arXiv (en), https://arxiv.org/abs/2609.22562 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7d6f1316cfac (Metraipolis archive dc_7d6f1316cfac, sha256 c0f2f27b2ea0abeb55a94ebab0fe248c6172da3d833e2685bf6c32697fd94c84)
BibTeX
@misc{dc7d6f1316cfac,
title = {AdaMerge: Tuning-Free Patch Compression for Multi-Vector Visual Document Retrieval — arXiv:2609.22562},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.22562},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7d6f1316cfac; sha256 c0f2f27b2ea0abeb55a94ebab0fe248c6172da3d833e2685bf6c32697fd94c84},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.22562. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Multi-vector visual document retrieval (VDR) models such as ColPali and ColNomic achieve strong accuracy by representing each document with hundreds to thousands of patch-level embeddings, at substantial storage and latency cost. Existing compression methods either prune unimportant patches or merge similar ones into clusters; the recent state-of-the-art merging method Prune-then-Merge (PtM) consistently outperforms pruning-only baselines at high compression, but requires a per-dataset cluster budget m to be tuned by grid search. We observe that the merge-cosine sequence produced by hierarchical clustering exhibits a sharp cliff separating mergeable redundancy from salient signal, and that the location of this cliff is concentrated in a narrow band across more than 11,000 documents from 14 datasets. This suggests the merge boundary can be detected per document rather than tuned per dataset. Building on this observation, we propose AdaMerge, a plug-and-play compression method that (i) detects each document's own cliff via gap analysis on the merge-cosine trajectory, and (ii) builds attention-weighted cluster centroids to preserve salient signal. On the long-document benchmark ViDoRe-V2 (4 datasets, two backbones), AdaMerge significantly outperforms tuned PtM across the operating range (p < 10^-4); on the short-document benchmark ViDoRe-V1 (10 datasets, two backbones), where all merging methods are already near-lossless, AdaMerge matches tuned PtM without any per-dataset tuning. AdaMerge adds only about 10 ms per document and exposes a single global hyperparameter shared across all datasets and backbones. Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-09-22 in cs.IR). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Information Retrieval (cs.IR). Pagina del lavoro: https://arxiv.org/abs/2609.22562 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-09T21:45:02.232Z dalla pagina https://arxiv.org/catchup/cs.IR/2026-09-22?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).