CNN-Transformer混合模型在计算机视觉领域的研究综述  

Review of CNN-Transformer Hybrid Model in Computer Vision

作　　者：戴洋毅 何康 瑚琦[1,2] 黄凯[1] 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海 [2]上海理工大学上海市现代光学系统重点实验室,上海

出　　处：《建模与仿真》2023年第4期3657-3672,共16页Modeling and Simulation

摘　　要：近年来,CNN-Transformer混合模型在计算机视觉领域的研究已经成为热点话题之一。这种模型可以结合卷积神经网络(Convolutional Neural Network, CNN)和Transformer各自的优势,提高模型在多种计算机视觉任务中的性能。首先对CNN与Transformer分别进行简述并分析其优缺点,然后通过介绍与分析近几年国内外表现出色的CNN-Transformer混合模型,对多种常见的混合方式进行分类阐述,这些方法旨在发挥卷积神经网络在局部特征提取方面的优势以及Transformer在全局信息建模方面的优势。最后,对CNN-Transformer混合模型在计算机视觉领域以及其他领域未来所面对的挑战和发展趋势进行展望。In recent years, research on CNN-Transformer hybrid models in computer vision has become one of the hottest topics. This type of model combines the advantages of Convolutional Neural Networks (CNN) and Transformers to improve the performance of various computer vision tasks. First, the pros and cons of CNN and Transformer are briefly introduced and analyzed. Subsequently, various common hybrid methods are elaborated through the introduction and analysis of outstanding CNN transformer hybrid models from national and international research in recent years. These meth-ods aim to leverage the local feature extraction capabilities of Convolutional Neural Networks and the global information modeling capabilities of Transformers. Finally, the paper looks at the chal-lenges and development trends facing CNN-Transformer hybrid models in computer vision and other fields in the future.

关 键 词：卷积神经网络 计算机视觉 局部特征提取 混合模型 CNN 混合方式 信息建模 热点话题 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]