Newest papers

DISTRACT: Degradation-Indexed Study of Textual Reference Across Competing Targets

Dan Zlotnikov, Ohad Ben‐Shahar · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28419 · cited by 0 works (OpenAlex) · open-access PDF

French

DISTRACT, un benchmark d’identification d’intervention correspondant à une cible, accepte des descriptions destinées à faciliter le suivi, mais l’acceptation du texte n’établit pas que son contenu modifie les décisions de suivi, en particulier lorsque les preuves visuelles se détériorent. Nous introduisons DISTRACT, un benchmark d’identification d’intervention correspondant qui maintient les vidéos, les cibles et les distracteurs fixes tout en traversant quatre priors de texte avec quatre dégradations d’image le long d’un axe de sévérité contrôlée. Nous évaluons cinq VLT sur quatre ensembles de données en utilisant des contrastes comportementaux, des sondes de représentation visuelle par couches et des patchs d’

Machine translation (Meta M2M100); the English original follows.

Abstract

Vision-language trackers (VLTs) accept descriptions intended to assist tracking, but accepting text does not establish that its content changes tracking decisions, particularly as visual evidence deteriorates. We introduce DISTRACT, a matched-intervention benchmark that holds video, target, and distractor fixed while crossing four text priors with four image degradations along a controlled severity axis. We evaluate five VLTs on four datasets using behavioral contrasts, layerwise representation probes, and activation patching. Molmo2 and JointNLT exhibit directed text use. Some effects contract under degradation, whereas others show no significant endpoint change after multiplicity correction. UVLTrack, MMTrack, and DUTrack respond only weakly to changes in text semantics, with no directed behavioral effect surviving FDR correction in the tested settings. Model-internal analyses reveal that at decision-near states, correct-versus-conflicting text is decodable in Molmo2 and JointNLT but not in the other three systems. In the responsive models, decodability peaks at late states where replacement produces little behavioral change, whereas replacing selected earlier states redirects conflict-associated trajectories toward the target. Because matched-correct donors do not significantly outperform unrelated donors, these interventions identify earlier intervention-sensitive states without establishing target-specific semantic rescue. DISTRACT thus exposes when and where text affects tracking under degraded visual evidence.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Eye tracking, Representation (politics), BitTorrent tracker, Artificial intelligence, Benchmark (surveying)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange