人工智能
计算机科学
对偶(语法数字)
领域(数学分析)
模式识别(心理学)
域适应
计算机视觉
适应(眼睛)
边距(机器学习)
数据挖掘
作者
Shenjian Gong,Yu Zhang,Rui Yang,Heng Yin,Xing Yang
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-04-21
卷期号:: 5796-5800
标识
DOI:10.1109/icassp55912.2026.11461444
摘要
Recently, RGBT crowd density estimation has garnered increasing interest. Its principal challenge is the high cost of achieving high-quality manual annotations for multi-modal data. To address this, we make the first attempt to introduce unsupervised domain adaptation (UDA) technique into RGBT counting to better transfer knowledge from easily accessible synthetic data to real-world scenarios. In this work, we first generate thermal infrared data from existing synthetic visible datasets via a pre-trained visible-to-infrared diffusion model. Subsequently, we introduce a dual alignment RGBT counting method (DAC), which comprises: (1) explicit data alignment (EDA) reduces domain discrepancies at data level through density/scene regularization, brightness adjustment, and perspective transformation; and (2) implicit feature alignment (IFA) aligns the modality-specific and multi-modal fusion features from different domains via adversarial learning, enabling the alleviation of feature level domain gap. Extensive experiments on real-world RGBT counting benchmark show the effectiveness of our method. The code is publicly at https://github.com/Yankeegsj/DAC.
科研通智能强力驱动
Strongly Powered by AbleSci AI