室内图像中家具多标签标注的实现

A Method of Dense Furniture Caption for Indoor Images

在基于图像进行家居虚拟设计的应用中,由于图像缺乏场景的深度信息、物体之间存在相互遮挡等问题,给获取图像信息带来一定的挑战。该文利用深度学习技术,提出了一种结合卷积神经网络和循环神经网络的方法,对室内图像进行特征提取,实现家具的多标签标注,以获取家具的属性信息,包括种类、位置、颜色和材质等。结果表明,该方法提高了虚拟展示内容的丰富性和精确性,为家居智能交互作了很好的铺垫。

In the application of image-based virtual house design system,it becomes a huge challenge to obtaining rich information from images because of some problems such as the lack of scene depth and the occlusion between objects.With the development of deep learning,this paper proposes a method of dense furniture caption for indoor images,which combines CNN and RNN to extract features.It can get multiple information of furniture,such as classification,location,color,material,etc.The result indicates that the method improves the richness and accuracy of furniture information,which makes a great contribution to virtual house design system.