双阶段框架与软标签蒸馏和空间提示图像-文本检索。

IF 2.6 3区 综合性期刊 Q1 MULTIDISCIPLINARY SCIENCES
PLoS ONE Pub Date : 2025-10-10 eCollection Date: 2025-01-01 DOI:10.1371/journal.pone.0333084
Ran Jin, Zhengang Li, Fang Deng, Yanhong Zhang, Min Luo, Tao Jin, Tengda Hou, Chenjie Du, Xiaozhe Gu, Jie Yuan
{"title":"双阶段框架与软标签蒸馏和空间提示图像-文本检索。","authors":"Ran Jin, Zhengang Li, Fang Deng, Yanhong Zhang, Min Luo, Tao Jin, Tengda Hou, Chenjie Du, Xiaozhe Gu, Jie Yuan","doi":"10.1371/journal.pone.0333084","DOIUrl":null,"url":null,"abstract":"<p><p>Vision-language pre-training (VLP) methods have significantly advanced cross-modal tasks in recent years. However, image-text retrieval still faces two critical challenges: inter-modal matching deficiency and intra-modal fine-grained localization deficiency. These issues significantly impede the accuracy of image-text retrieval. To address these challenges, we propose a novel dual-stage training framework. In the first stage, we employ Soft Label Distillation (SLD) to align the contrastive relationships between images and texts by mitigating the overfitting problem caused by hard labels. In the second stage, we introduce Spatial Text Prompt (STP) to enhance the model's visual grounding capabilities by incorporating spatial prompt information, thereby achieving more precise fine-grained alignment. Extensive experiments on standard datasets show that our method outperforms state-of-the-art approaches in image-text retrieval.The code and supplementary files can be found at https://github.com/Leon001211/DSSLP.</p>","PeriodicalId":20189,"journal":{"name":"PLoS ONE","volume":"20 10","pages":"e0333084"},"PeriodicalIF":2.6000,"publicationDate":"2025-10-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12513663/pdf/","citationCount":"0","resultStr":"{\"title\":\"Dual-stage framework with soft-label distillation and spatial prompting for image-text retrieval.\",\"authors\":\"Ran Jin, Zhengang Li, Fang Deng, Yanhong Zhang, Min Luo, Tao Jin, Tengda Hou, Chenjie Du, Xiaozhe Gu, Jie Yuan\",\"doi\":\"10.1371/journal.pone.0333084\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>Vision-language pre-training (VLP) methods have significantly advanced cross-modal tasks in recent years. However, image-text retrieval still faces two critical challenges: inter-modal matching deficiency and intra-modal fine-grained localization deficiency. These issues significantly impede the accuracy of image-text retrieval. To address these challenges, we propose a novel dual-stage training framework. In the first stage, we employ Soft Label Distillation (SLD) to align the contrastive relationships between images and texts by mitigating the overfitting problem caused by hard labels. In the second stage, we introduce Spatial Text Prompt (STP) to enhance the model's visual grounding capabilities by incorporating spatial prompt information, thereby achieving more precise fine-grained alignment. Extensive experiments on standard datasets show that our method outperforms state-of-the-art approaches in image-text retrieval.The code and supplementary files can be found at https://github.com/Leon001211/DSSLP.</p>\",\"PeriodicalId\":20189,\"journal\":{\"name\":\"PLoS ONE\",\"volume\":\"20 10\",\"pages\":\"e0333084\"},\"PeriodicalIF\":2.6000,\"publicationDate\":\"2025-10-10\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12513663/pdf/\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"PLoS ONE\",\"FirstCategoryId\":\"103\",\"ListUrlMain\":\"https://doi.org/10.1371/journal.pone.0333084\",\"RegionNum\":3,\"RegionCategory\":\"综合性期刊\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"2025/1/1 0:00:00\",\"PubModel\":\"eCollection\",\"JCR\":\"Q1\",\"JCRName\":\"MULTIDISCIPLINARY SCIENCES\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"PLoS ONE","FirstCategoryId":"103","ListUrlMain":"https://doi.org/10.1371/journal.pone.0333084","RegionNum":3,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2025/1/1 0:00:00","PubModel":"eCollection","JCR":"Q1","JCRName":"MULTIDISCIPLINARY SCIENCES","Score":null,"Total":0}
引用次数: 0

摘要

近年来,视觉语言预训练(VLP)方法在跨模态任务方面取得了显著进展。然而,图像-文本检索仍然面临着两大关键挑战:模态间匹配不足和模态内细粒度定位不足。这些问题严重影响了图像-文本检索的准确性。为了应对这些挑战,我们提出了一种新的双阶段培训框架。在第一阶段,我们使用软标签蒸馏(SLD)通过减轻硬标签引起的过拟合问题来对齐图像和文本之间的对比关系。在第二阶段,我们引入空间文本提示(STP),通过整合空间提示信息来增强模型的视觉接地能力,从而实现更精确的细粒度对齐。在标准数据集上进行的大量实验表明,我们的方法在图像-文本检索方面优于最先进的方法。代码和补充文件可以在https://github.com/Leon001211/DSSLP上找到。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Dual-stage framework with soft-label distillation and spatial prompting for image-text retrieval.

Vision-language pre-training (VLP) methods have significantly advanced cross-modal tasks in recent years. However, image-text retrieval still faces two critical challenges: inter-modal matching deficiency and intra-modal fine-grained localization deficiency. These issues significantly impede the accuracy of image-text retrieval. To address these challenges, we propose a novel dual-stage training framework. In the first stage, we employ Soft Label Distillation (SLD) to align the contrastive relationships between images and texts by mitigating the overfitting problem caused by hard labels. In the second stage, we introduce Spatial Text Prompt (STP) to enhance the model's visual grounding capabilities by incorporating spatial prompt information, thereby achieving more precise fine-grained alignment. Extensive experiments on standard datasets show that our method outperforms state-of-the-art approaches in image-text retrieval.The code and supplementary files can be found at https://github.com/Leon001211/DSSLP.

求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
PLoS ONE
PLoS ONE 生物-生物学
CiteScore
6.20
自引率
5.40%
发文量
14242
审稿时长
3.7 months
期刊介绍: PLOS ONE is an international, peer-reviewed, open-access, online publication. PLOS ONE welcomes reports on primary research from any scientific discipline. It provides: * Open-access—freely accessible online, authors retain copyright * Fast publication times * Peer review by expert, practicing researchers * Post-publication tools to indicate quality and impact * Community-based dialogue on articles * Worldwide media coverage
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信