人工智能
计算机视觉
计算机科学
跟踪(教育)
融合
传感器融合
噪音(视频)
职位(财务)
跟踪系统
模式识别(心理学)
算法
作者
Ze Zhou,Chen Wu,YongWu,Chaobo Li,Yanwei Xu
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-04-21
卷期号:: 10382-10386
标识
DOI:10.1109/icassp55912.2026.11464901
摘要
Incorporating linguistic information opens new opportunities for visual object tracking. However, existing vision-language trackers still suffer from limited cross-modal interaction, making it challenging to achieve deep fusion between visual and textual representations and to fully exploit textual cues throughout the entire tracking process. To address these issues, we propose DIFTrack, a novel Deep Interaction Fusion Tracker that efficiently and comprehensively fuses and utilizes textual information. Specifically, in the feature fusion stage, we design a Bidirectional Interaction Module (BIM) that leverages bidirectional attention to enable deep interaction between visual and textual features, yielding more discriminative cross-modal representations. Furthermore, in the feature decoding stage, we introduce a Text-Guided Module (TGM) that continuously injects text semantic information into multi-cue queries, fully harnessing textual guidance during the decoding process. Extensive experiments on multiple benchmarks including TNL2K, LaSOT, and LaSOTextdemonstrate that our approach achieves comparable performance to state-of-the-art trackers. The source code will be released at https://github.com/ChenWu6369/DIFTrack.
科研通智能强力驱动
Strongly Powered by AbleSci AI