基于深度学习的视觉关系检测方法及应用  

Method and Application of Visual Relationship Detection Based on Deep Learning

作　　者：汤婧婧 黄晶[2] 石爱业[1] 张丽丽[1] 徐立中[1] 

机构地区：[1]河海大学计算机与信息学院,江苏 南京 [2]河海大学商学院,江苏 南京

出　　处：《图像与信号处理》2022年第3期144-161,共18页Journal of Image and Signal Processing

摘　　要：随着深度学习的不断发展和广泛应用,计算机视觉的许多领域也得到了长足的进步,例如在图像分类、对象检测、图像分割等任务中的表现。视觉关系检测(VRD)是计算机视觉的重要任务,旨在识别图像中物体之间的关系或相互作用,这对于理解图像及视觉世界都很重要,VRD也是计算机视觉技术应用研究的关键环节。与一般的物体检测任务相比,VRD不仅需要预测每个物体的类别和轨迹,还需要预测物体之间的关系,研究人员已经针对改任务提出了很多办法,特别在近年来基于深度神经网络的发展的深度学习也有所突破。本文介绍了VRD任务的内容,深度学习基本方法,VRD的传统方法和基于深度学习模型的一些分类和框架及其VRD在计算机视觉领域的应用。With the continuous development and wide application of deep learning, many fields of computer vision have also made great progress, such as performance in image classification, object detection, image segmentation and other tasks. Visual relationship detection (VRD) is an important task for computer vision, aiming to recognize relations or interactions between objects in an image, which is important for understanding images even the visual world. Compared with the general object detection task, VRD requires not only to predict the categories and trajectories of each object, but also to predict the relationship between objects. Researchers have proposed to tackle this problem especially with the development of deep neural networks in recent years. In this survey, we provide a comprehensive review of VRD in computer vision and some categorization and frameworks of deep learning models for VRD with its applications.

关 键 词：计算机视觉 深度学习 神经网络 视觉关系检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]