跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.23432 · v1

Bits, Not Boxes: Label-Free Layer Selection for Visual Localization in VLMs

首次提交 2026-11-18 03:37
版本更新 2026-11-18 03:37
雷达观测 2026-11-19 19:08
arXiv ↗PDF ↗

摘要

Attention maps cheaply show where a vision-language model (VLM) looks, but attention sinks make them unreliable for localization. Contrasting the question with a generic prompt suppresses sinks, yet only some layers localize the queried object. Existing methods find these layers with box annotations or fixed rules. We show that attention itself can identify them. A question changes how much a head attends to the image and where this attention goes. The two effects separate exactly, and the second, measured in bits on unlabeled questions, indicates which layers localize. WhereBits turns these where-to-look bits into a layer prior and combines it with per-head contrastive scores. Its three design constants were fixed once on labeled COCO data, and a new model or resolution needs only unlabeled questions. Across eight VLMs, WhereBits matches the pointing accuracy of box-tuned layers on COCO and RefCOCO and ranks target tokens better. This advantage holds on a held-out model family and at higher resolutions. Its COCO pointing accuracy is 0.72, against 0.22 for mean attention. In training-free zoom-in, its crops raise V* accuracy by 6.0 points under forced-choice evaluation.

作者备注:35 pages, 9 figures, 56 tables

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-18 03:37 · 最新观测版本

Bits, Not Boxes: Label-Free Layer Selection for Visual Localization in VLMs

Attention maps cheaply show where a vision-language model (VLM) looks, but attention sinks make them unreliable for localization. Contrasting the question with a generic prompt suppresses sinks, yet only some layers localize the queried object. Existing methods find these layers with box annotations or fixed rules. We show that attention itself can identify them. A question changes how much a head attends to the image and where this attention goes. The two effects separate exactly, and the second, measured in bits on unlabeled questions, indicates which layers localize. WhereBits turns these where-to-look bits into a layer prior and combines it with per-head contrastive scores. Its three design constants were fixed once on labeled COCO data, and a new model or resolution needs only unlabeled questions. Across eight VLMs, WhereBits matches the pointing accuracy of box-tuned layers on COCO and RefCOCO and ranks target tokens better. This advantage holds on a held-out model family and at higher resolutions. Its COCO pointing accuracy is 0.72, against 0.22 for mean attention. In training-free zoom-in, its crops raise V* accuracy by 6.0 points under forced-choice evaluation.

查看此版本解读 →查看此版本原文 ↗