Metraipolis › archive › dc_742a567405cf

Look Back, Think Ahead: Visual Memory on Demand for Efficient Multimodal Reasoning — arXiv:2610.12060

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.12060
License (as archived)
CC0-1.0
Archived
2026-10-09T06:30:47.446Z
SHA-256 of the archived text
f3877b525917d36f0dd079f4894b37ef1a2048145d171f5de618554778f8b33c
Stable URL of this page
https://metraipolis.com/documento/dc_742a567405cf

How to cite

Look Back, Think Ahead: Visual Memory on Demand for Efficient Multimodal Reasoning — arXiv:2610.12060 — arXiv (en), https://arxiv.org/abs/2610.12060 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_742a567405cf (Metraipolis archive dc_742a567405cf, sha256 f3877b525917d36f0dd079f4894b37ef1a2048145d171f5de618554778f8b33c)
BibTeX
@misc{dc742a567405cf,
  title = {Look Back, Think Ahead: Visual Memory on Demand for Efficient Multimodal Reasoning — arXiv:2610.12060},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.12060},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_742a567405cf; sha256 f3877b525917d36f0dd079f4894b37ef1a2048145d171f5de618554778f8b33c},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.12060. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Processing long visual token sequences from high-resolution images makes multi-step reasoning computationally expensive for multimodal Large Language Models (MLLMs). Existing one-shot pruning and aggregation methods compress visual tokens into a fixed context before decoding. However, visual evidence needs can shift as reasoning unfolds, making it difficult for a fixed compressed context to retain all the details needed across stages. To address this challenge, we propose ViMoD, a lightweight framework that maintains a compact visual context while preserving access to original fine-grained evidence as reasoning needs evolve. Deformable Aggregation of Region-wise Tokens (DART) learns content-adaptive groups and aggregation capacities, constructing compact Coarse representations linked to recoverable original Fine tokens. Temporal Routing for Adaptive Contextual Evidence (TRACE) integrates decoding history to anticipate upcoming evidence needs and select, retain, or replace active Fine-token groups. Selected Fine tokens augment the persistent Coarse context in the frozen backbone, enabling stage-specific evidence access without continuously attending to all visual tokens. On Qwen3-VL-4B, ViMoD outperforms all evaluated baselines on all eight reasoning benchmarks at a 20% target visual token budget, improving the mean normalized score by 39.0% over the strongest evaluated one-shot baseline. These gains are achieved with only 0.0546% additional trainable parameters relative to the frozen backbone.

Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.CV, nuovi lavori del giorno).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2610.12060 — 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 2026-10-09T05:55:07.604Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).