光电技术应用, 2020, 35 (2): 60, 网络出版: 2020-05-28
基于无监督学习的单人多姿态图像生成方法
Single-personmulti-pose Image Generation Method Based on Unsupervised Learning
半监督学习 视觉监控 生成对抗网络 结构相似度 semi-supervised learning visual surveillance generative adversarial network structural similari. ty index
摘要
针对目前视觉监控领域中采集到的人物数据样本量少和特征单一的问题,提出了一种具有高视觉感知约束的双向生成对抗网络生成期望人物姿态图像的方法。采用给定人物的单个图像和期望姿态的二维骨架作为双向生成对抗网络的输入,生成具有该目标人物期望姿态的图像。将生成的期望姿态图像反映射回原始姿态图像,利用少量的图像以无监督学习方式进行学习,生成该人物期望姿态的高质量图像。提出的方法在 DeepFashion公开数据集上进行了实验,结果表明,采用文中提出的方法生成的图像结构相似度(SSIM)比以往的方法提高了 0.28,有效的提升了基于无监督学习的单人多姿态人物图像生成的质量。
Abstract
Aiming at the problems of small sample data and single feature of the person data collected in thefield of visual surveillance, a method of generating a desired person pose image with a bidirectional generative ad.versarial network with high visual perception constraints is proposed. A single image of a given character and atwo-dimensional skeleton of a desired pose are used as inputs to a bidirectional generation adversarial network togenerate an image with the desired pose of the target person. The generated expected pose image is mapped back tothe original pose image, and a small number of images are used for learning in an unsupervised learning manner togenerate a high-quality image of the character′s desired pose. The proposed method is tested on the DeepFashionpublic data set. The results show that the image structure similarity (SSIM) generated by the method is 0.28 higherthan that of previous methods, which effectively improves the image generation quality of single-personmulti-pose based on unsupervised learning.
张婧, 孙金根, 陈亮, 刘韵婷. 基于无监督学习的单人多姿态图像生成方法[J]. 光电技术应用, 2020, 35(2): 60. ZHANG Jing, SUN Jin-gen, CHEN Liang, LIU Yun-ting. Single-personmulti-pose Image Generation Method Based on Unsupervised Learning[J]. Electro-Optic Technology Application, 2020, 35(2): 60.