Metraipolis › archive › dc_e94bf63eb86c
DIVE: Dynamic Iterative Visual Evidence Construction for Efficient Vision-Language Models — arXiv:2608.04496
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.04496
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-31T11:35:42.674Z
- SHA-256 of the archived text
1dfd9966c040fafa4b1876691ffd9c189004eda240af24a46e053063bb01c866- Stable URL of this page
https://metraipolis.com/documento/dc_e94bf63eb86c
How to cite
DIVE: Dynamic Iterative Visual Evidence Construction for Efficient Vision-Language Models — arXiv:2608.04496 — arXiv (en), https://arxiv.org/abs/2608.04496 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e94bf63eb86c (Metraipolis archive dc_e94bf63eb86c, sha256 1dfd9966c040fafa4b1876691ffd9c189004eda240af24a46e053063bb01c866)
BibTeX
@misc{dce94bf63eb86c,
title = {DIVE: Dynamic Iterative Visual Evidence Construction for Efficient Vision-Language Models — arXiv:2608.04496},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.04496},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e94bf63eb86c; sha256 1dfd9966c040fafa4b1876691ffd9c189004eda240af24a46e053063bb01c866},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.04496. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Visual inputs in vision-language models (VLMs) are often encoded into substantially longer token sequences than text, making visual tokens a major bottleneck for efficient inference. Abundant recent methods address this bottleneck by scoring token importance and pruning low-scoring tokens in a single pass. However, one-shot scoring is insufficient because a token's prompt-relevant usefulness depends on the evidence already retained. Motivated by this insight, we introduce DIVE (Dynamic Iterative Visual Evidence Construction), a training-free framework that recasts visual-token pruning as dynamic evidence construction. DIVE repeatedly selects the remaining token with the highest residual-conditioned score, updates the visual and prompt residuals to discount the evidence already explained, and re-evaluates the remaining tokens. This select-update-re-evaluate process builds a retained set of complementary, prompt-relevant evidence. Experiments across eight image-understanding benchmarks show that DIVE consistently preserves performance across token budgets. With an 88.9% reduction in visual tokens, DIVE retains 98.2% of the uncompressed model's average performance. Code is available at this https URL . Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-06 in cs.LG). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2608.04496 — 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 2026-10-31T11:24:48.833Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-08-06?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).