目的随着电影内容的复杂化与多样化,电影场景分割成为理解影片结构和支持多媒体应用的重要任务。为提升镜头特征提取和特征关联的有效性,增强镜头序列的上下文感知能力,提出一种混合架构电影场景分割方法(hybrid architecture scene seg...目的随着电影内容的复杂化与多样化,电影场景分割成为理解影片结构和支持多媒体应用的重要任务。为提升镜头特征提取和特征关联的有效性,增强镜头序列的上下文感知能力,提出一种混合架构电影场景分割方法(hybrid architecture scene segmentation network,HASSNet)。方法首先,采用预训练结合微调策略,在大量无场景标签的电影数据上进行无监督预训练,使模型学习有效的镜头特征表示和关联特性,然后在有场景标签的数据上进行微调训练,进一步提升模型性能;其次,模型架构上混合了状态空间模型和自注意力机制模型,分别设计Shot Mamba镜头特征提取模块和Scene Transformer特征关联模块,Shot Mamba通过对镜头图像分块建模提取有效特征表示,Scene Transformer则通过注意力机制对不同镜头特征进行关联建模;最后,采用3种无监督损失函数进行预训练,提升模型在镜头特征提取和关联上的性能,并使用Focal Loss损失函数进行微调,以改善由于类别不平衡导致的精度不足问题。结果实验结果表明,HASSNet在3个数据集上显著提升了场景分割的精度,在典型电影场景分割数据集MovieNet中,与先进的场景分割方法相比,AP(average precision)、mIoU(mean intersection over union)、AUC-ROC(area under the receiver operating characteristic curve)和F1分别提升1.66%、10.54%、0.21%和16.83%,验证了本文提出的HASSNet方法可以有效提升场景边界定位的准确性。结论本文提出的HASSNet方法有效结合了预训练与微调策略,借助混合状态空间模型和自注意力机制模型的特点,增强了镜头的上下文感知能力,使电影场景分割的结果更加准确。展开更多
In the era of information and communication technology (ICT) and big data, the map gradually shows a new qualitative feature of “spatiotemporal ubiquitous” with the extension of its object space, expression space an...In the era of information and communication technology (ICT) and big data, the map gradually shows a new qualitative feature of “spatiotemporal ubiquitous” with the extension of its object space, expression space and information source, which challenges the theory of cartographic visualization. This paper discusses the ubiquitous map visualization from the object content and expression form. Oriented to the ternary space, it divides the object dimension of ubiquitous map visualization and analyzes the expression characteristics of ubiquitous map visualization. Based on that, it constructs the variable system, symbol system and method system of ubiquitous map visualization. With three cases of the metro roadmap, the tag map, and the three-dimensional (3D) city map, the application of the proposed content is explained to illustrate its effectiveness. The research in this paper is expected to further enrich the theoretical basis of cartographic visualization and provide theoretical support for the expression and application of ubiquitous map visualization.展开更多
文摘目的随着电影内容的复杂化与多样化,电影场景分割成为理解影片结构和支持多媒体应用的重要任务。为提升镜头特征提取和特征关联的有效性,增强镜头序列的上下文感知能力,提出一种混合架构电影场景分割方法(hybrid architecture scene segmentation network,HASSNet)。方法首先,采用预训练结合微调策略,在大量无场景标签的电影数据上进行无监督预训练,使模型学习有效的镜头特征表示和关联特性,然后在有场景标签的数据上进行微调训练,进一步提升模型性能;其次,模型架构上混合了状态空间模型和自注意力机制模型,分别设计Shot Mamba镜头特征提取模块和Scene Transformer特征关联模块,Shot Mamba通过对镜头图像分块建模提取有效特征表示,Scene Transformer则通过注意力机制对不同镜头特征进行关联建模;最后,采用3种无监督损失函数进行预训练,提升模型在镜头特征提取和关联上的性能,并使用Focal Loss损失函数进行微调,以改善由于类别不平衡导致的精度不足问题。结果实验结果表明,HASSNet在3个数据集上显著提升了场景分割的精度,在典型电影场景分割数据集MovieNet中,与先进的场景分割方法相比,AP(average precision)、mIoU(mean intersection over union)、AUC-ROC(area under the receiver operating characteristic curve)和F1分别提升1.66%、10.54%、0.21%和16.83%,验证了本文提出的HASSNet方法可以有效提升场景边界定位的准确性。结论本文提出的HASSNet方法有效结合了预训练与微调策略,借助混合状态空间模型和自注意力机制模型的特点,增强了镜头的上下文感知能力,使电影场景分割的结果更加准确。
文摘In the era of information and communication technology (ICT) and big data, the map gradually shows a new qualitative feature of “spatiotemporal ubiquitous” with the extension of its object space, expression space and information source, which challenges the theory of cartographic visualization. This paper discusses the ubiquitous map visualization from the object content and expression form. Oriented to the ternary space, it divides the object dimension of ubiquitous map visualization and analyzes the expression characteristics of ubiquitous map visualization. Based on that, it constructs the variable system, symbol system and method system of ubiquitous map visualization. With three cases of the metro roadmap, the tag map, and the three-dimensional (3D) city map, the application of the proposed content is explained to illustrate its effectiveness. The research in this paper is expected to further enrich the theoretical basis of cartographic visualization and provide theoretical support for the expression and application of ubiquitous map visualization.
