Metraipolis › archive › dc_46a32e95cc3d

StackTok: Accelerating VLMs Inference with Budget-Adaptive Visual Token Selection — arXiv:2609.16841

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.16841
License (as archived)
CC0-1.0
Archived
2026-10-15T18:38:40.880Z
SHA-256 of the archived text
5e8a12f36d75dd0e8bf7bb5676855b8ac7abb1bf07ad9435770d0c4c6b632488
Stable URL of this page
https://metraipolis.com/documento/dc_46a32e95cc3d

How to cite

StackTok: Accelerating VLMs Inference with Budget-Adaptive Visual Token Selection — arXiv:2609.16841 — arXiv (en), https://arxiv.org/abs/2609.16841 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_46a32e95cc3d (Metraipolis archive dc_46a32e95cc3d, sha256 5e8a12f36d75dd0e8bf7bb5676855b8ac7abb1bf07ad9435770d0c4c6b632488)
BibTeX
@misc{dc46a32e95cc3d,
  title = {StackTok: Accelerating VLMs Inference with Budget-Adaptive Visual Token Selection — arXiv:2609.16841},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.16841},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_46a32e95cc3d; sha256 5e8a12f36d75dd0e8bf7bb5676855b8ac7abb1bf07ad9435770d0c4c6b632488},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.16841. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Increasing image resolution produces ever-longer visual-token sequences in vision-language models (VLMs), substantially raising their inference cost. To reduce this overhead without retraining, existing methods select compact token subsets that prioritize query relevance, visual coverage, or a fixed trade-off between them. The appropriate balance, however, varies across queries and token budgets: localized questions favor relevance, whereas holistic questions demand broader visual coverage. We introduce StackTok, a training-free selector that treats query relevance as the objective and visual coverage as budget-calibrated support. StackTok builds a size-indexed coverage reference from a coverage-only greedy sequence and adjusts its support target using query--vision affinity entropy. A reference-gated interleaved selection policy then switches between relevance- and coverage-oriented additions according to the current subset's support deficit. For high-resolution inputs, StackTok allocates one shared token budget across crops according to the combined marginal gain of locally nominated tokens. Evaluated with five VLMs over ten distinct image-understanding benchmarks, StackTok ranks first among training-free selectors in every tested model--budget setting. On high-resolution LLaVA-NeXT-7B, it retains 95.26% of full-token performance with only 160 of 2{,}880 (5.6%) visual tokens.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-16 in cs.AI).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.16841 — 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 2026-10-15T18:03:09.694Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-16?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).