analiza aiqbench
ProtoLIP introduces a lightweight mechanism to disentangle visual evidence in vision-language models, separating object-specific signals from shared context without requiring spatial annotations or backbone retraining. By organizing reusable visual prototypes into semantic families, the method improves localization accuracy and transfers effectively to independently pretrained architectures using only text-derived weak supervision. This advancement addresses a key interpretability challenge in query-conditioned VLMs by enabling exact decomposition of matching scores into semantic and prototype contributions.
ProtoLIP: From Sentence-Level to Object-Level Evidence Disentanglement
arXiv:2609.16284v1 Announce Type: cross Abstract: Query-conditioned vision--language models enable fine-grained interpretation by revealing how visual evidence changes with textual queries. However, evidence conditioned on complete descriptions does not necessarily resolve into object-specific evidence, nor does an exposed evidence map necessarily identify the evidence that constitutes the model's prediction. Across multiple VLM architectures and independent benchmarks, we find that object-level queries often retain evidence from co-occurring objects and shared context. In this paper, we introduce \textbf{ProtoLIP}, a lightweight prototype-mediated evidence layer that organizes reusable visual prototypes into text-derived semantic families and uses query-dependent family routing to constrain which prototypes may provide evidence. Without spatial annotations or backbone retraining, ProtoLIP improves evidence localization and separation across query granularities, with localization gains transferring to independently pretrained VLMs with well-aligned patch--text representations. Despite using only text-derived weak supervision, ProtoLIP remains competitive with a spatially supervised grounding model while maintaining strong matching and competitive image--text retrieval. Crucially, ProtoLIP constructs its matching score directly from localized prototype evidence, enabling the score to be exactly decomposed into semantic-family and prototype contributions.
Abstract:Query-conditioned vision--language models enable fine-grained interpretation by revealing how visual evidence changes with textual queries. However, evidence conditioned on complete descriptions does not necessarily resolve into object-specific evidence, nor does an exposed evidence map necessarily identify the evidence that constitutes the model's prediction. Across multiple VLM architectures and independent benchmarks, we find that object-level queries often retain evidence from co-occurring objects and shared context. In this paper, we introduce \textbf{ProtoLIP}, a lightweight prototype-mediated evidence layer that organizes reusable visual prototypes into text-derived semantic families and uses query-dependent family routing to constrain which prototypes may provide evidence. Without spatial annotations or backbone retraining, ProtoLIP improves evidence localization and separation across query granularities, with localization gains transferring to independently pretrained VLMs with well-aligned patch--text representations. Despite using only text-derived weak supervision, ProtoLIP remains competitive with a spatially supervised grounding model while maintaining strong matching and competitive image--text retrieval. Crucially, ProtoLIP constructs its matching score directly from localized prototype evidence, enabling the score to be exactly decomposed into semantic-family and prototype contributions.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2609.16284 [cs.CV]
(or arXiv:2609.16284v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.16284 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 269 cuvinte
Subiecte conexe
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026
Neural networks are not already clean
LessWrong · 17 oct. 2026
LiveLaw Supreme Court Half-Yearly Digest: January - June 2026 [Part 2]
livelaw.in · 13 sept. 2026
Implementing and Evaluating a Basic Per-Action Monitor for Safer Evals
METR · 27 sept. 2026