计算机科学
对抗制
人工智能
散列函数
钥匙(锁)
素描
方案(数学)
特征(语言学)
边距(机器学习)
自然语言处理
作者
Huan Liang,Xiaohui Wan,Ni He
出处
期刊:
日期:2026-07-01
卷期号:: 1-6
标识
DOI:10.1109/csrswtc71154.2026.11669098
摘要
To enhance the consistency of multi-modal representations and improve cross-modal retrieval performance, we propose an improved cross-modal hashing retrieval method that integrates CLIP with adversarial learning, aiming to build a more robust multi-modal retrieval model. Specifically, the CLIP model is employed to extract features from images and texts, thereby enhancing multi-modal semantic alignment. Adversarial learning is incorporated into both feature processing and hashing learning modules to generate discriminative and consistent hash codes. Additionally, angular triplet loss and classification loss are used to preserve the original data structure. Experiments on three benchmark datasets demonstrate the effectiveness of the proposed model.
科研通智能强力驱动
Strongly Powered by AbleSci AI