Papers/2611.23432
🧪 Test?View on arXiv

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

Author1, Author2, Author3, Author4, Author5

multimodallocalizationattention mechanisms
2611.23432
Builder Relevance
80%
15h ago

Abstract

This paper presents a method for identifying layers in vision-language models that localize objects without the need for box annotations.

Reality Card

Core Claim

The WhereBits method achieves pointing accuracy of 0.72 on COCO, matching box-tuned layers while requiring only unlabeled questions.

Method / Result

WhereBits improves V* accuracy by 6.0 points in training-free zoom-in evaluations.

Limitations

The method's constants were fixed on labeled COCO data, which may limit generalizability to other datasets.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers