arXiv cs.AI· Rahim Hossain, Dipanwita Saha, Ashirbad Mishra, Xinxin Shu, Svetlana Strunjas, Binbin Li·· 4 小时前AI 评分29
LoCaR:用冻结视觉语言模型实现局部画布检索
LoCaR: Localized Canvas Retrieval With Frozen Vision-Language Model
AI 导读
LoCaR 是一个无需训练的局部文本到图像检索框架,借助 SAM2 构建独立候选区域,用冻结视觉编码器分别编码每个区域以获得更局部的表示。研究还提出可选的 paired-support 适配,利用匹配的文本区域样例作为残差,且不更新底层编码器。
正文 · 原文
Abstract:Whole image embeddings can under represent small objects and local attributes that distinguish relevant images for localized queries. We present LoCaR, a training free framework for localized text to image retrieval with frozen vision language models. LoCaR constructs and utilizes independent proposals (SAM2) focusing on the object so as to get a more localized representation that independently encodes each region using the frozen visual encoder. We further study paired-support adaptation as an optional residual that uses matched text region exemplars without updating the underlying encoders. We also introduce LOCFG, a Visual Genome derived benchmark designed to evaluate localized retrieval for smaller and more challenging referents. Across several vision language backbones, our results show that improving the region representation itself is the main source of gain.
| Comments: | Zero-shot retrieval , Local Retrieval |
| Subjects: | Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) |
| Cite as: | arXiv:2611.22131 [cs.AI] |
| (or arXiv:2611.22131v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.22131 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Rahim Hossain [view email]
[v1]
Tue, 17 Nov 2026 03:15:02 UTC (10,231 KB)
来源:arXiv cs.AI · arxiv.org