李泽君_复旦大 学_An Unsupervised Sampling Approach for Image-Sentence Matching Using Document-level Structural Information_PPT.pdf
10墨值下载
An Unsupervised Sampling
Approach for Image-Sentence
Matching Using
Document-Level Structural
Information
Zejun Li
1
, Zhongyu Wei
1
, Zhihao Fan
1
, Haijun Shan
2
, Xuanjing
Huang
3
1
School of Data Science, Fudan University, China
2
Zhejiang Lab, China
3
School of Computer Science, Fudan University, China
Motivation
• Cross-modality Alignment:
• Alignment of 2 semantic spaces
(a)
• Image-sentence matching
• Contrastive Learning:
• Positive: Matched pairs
• Negative: Hard negative
• Information is needed to find
positive/negative pairs:
• Supervised: labeled pairs (b)
• Unsupervised
of 19
10墨值下载
【版权声明】本文为墨天轮用户原创内容，转载时必须标注文档的来源（墨天轮），文档链接，文档作者等基本信息，否则作者和墨天轮有权追究责任。如果您发现墨天轮中有涉嫌抄袭或者侵权的内容，欢迎发送邮件至：contact@modb.pro进行举报，并提供相关证据，一经查实，墨天轮将立刻删除相关内容。
下载排行榜
文档被以下合辑收录
评论