Metraipolis › archive › dc_0fa8b84e4772
Seeing is Believing: Benchmarking and Mitigating the Interference of Language Priors on Visual Perception in MLLMs — arXiv:2610.21902
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.21902
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T10:35:32.510Z
- SHA-256 of the archived text
a119ad150b03a4032b6586c30698e2d46fd4d39ff9df659e5ae893b10c2b79f9- Stable URL of this page
https://metraipolis.com/documento/dc_0fa8b84e4772
How to cite
Seeing is Believing: Benchmarking and Mitigating the Interference of Language Priors on Visual Perception in MLLMs — arXiv:2610.21902 — arXiv (en), https://arxiv.org/abs/2610.21902 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0fa8b84e4772 (Metraipolis archive dc_0fa8b84e4772, sha256 a119ad150b03a4032b6586c30698e2d46fd4d39ff9df659e5ae893b10c2b79f9)
BibTeX
@misc{dc0fa8b84e4772,
title = {Seeing is Believing: Benchmarking and Mitigating the Interference of Language Priors on Visual Perception in MLLMs — arXiv:2610.21902},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.21902},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0fa8b84e4772; sha256 a119ad150b03a4032b6586c30698e2d46fd4d39ff9df659e5ae893b10c2b79f9},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.21902. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Language priors in MLLMs can override visual evidence, impairing visual perception. Existing benchmarks assume shared priors across models and rely solely on accuracy. However, cross-model prior agreement falls below 80%, and over 15% of items can be answered correctly without any image, causing accuracy to overestimate prior resistance. To address these limitations, we introduce ViSiP-Bench, a benchmark grounded in rare-but-real scenarios across six prior categories, with a model-aware dynamic prior induction protocol and three metrics that more precisely and continuously measure prior resistance than accuracy alone. Evaluating 24 models reveals that prior adherence is pervasive, does not improve with scale, and larger models tend to either strongly override their priors or firmly adhere to them. We therefore propose MAPS, which identifies each model's strongest priors and fine-tunes on factually grounded knowledge to soften excessive prior reliance without disrupting general capabilities. 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/2610.21902 — 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 2026-10-16T08:52:02.706Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).