arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2508.06853cs.CVcs.AI

AGIC:注意力引导的图像描述生成以提升描述相关性

AGIC: Attention-Guided Image Captioning to Improve Caption Relevance

NIT Silchar(NIT西里char);TCS Research, Hyderabad(TCS研究 Hyderabad);SVNIT Surat, India(SVNIT Surat印度)

L. D. M. S. Sai Teja, Ashok Urlana, Pruthwik Mishra

AI总结:

本文提出AGIC框架,通过在特征空间放大显著视觉区域引导描述生成,并引入混合解码策略平衡流畅性与多样性,在Flickr8k和Flickr30k上实验表明其性能超越多个先进模型且推理更快。

摘要:

尽管图像描述生成取得了显著进展,但生成准确且具描述性的描述仍是一项长期挑战。在本研究中,我们提出了Attention-Guided Image Captioning(AGIC,注意力引导的图像描述生成),该方法直接在特征空间中放大显著视觉区域以引导描述生成。我们进一步引入了一种结合确定性采样和概率采样的混合解码策略,以平衡流畅性和多样性。为了评估AGIC,我们在Flickr8k和Flickr30k数据集上进行了大量实验。结果表明,AGIC匹配或超越了几种最先进的模型,同时实现了更快的推理。此外,AGIC在多个评估指标上表现出色,为图像描述生成提供了一种可扩展且可解释的解决方案。

英文摘要:

Despite significant progress in image captioning, generating accurate and descriptive captions remains a long-standing challenge. In this study, we propose Attention-Guided Image Captioning (AGIC), which amplifies salient visual regions directly in the feature space to guide caption generation. We further introduce a hybrid decoding strategy that combines deterministic and probabilistic sampling to balance fluency and diversity. To evaluate AGIC, we conduct extensive experiments on the Flickr8k and Flickr30k datasets. The results show that AGIC matches or surpasses several state-of-the-art models while achieving faster inference. Moreover, AGIC demonstrates strong performance across multiple evaluation metrics, offering a scalable and interpretable solution for image captioning.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑