Investigation of Facial Preference Using Gaussian Process Preference Learning and Generative Image Model
Investigation of Facial Preference Using Gaussian Process Preference Learning and Generative Image Model
复制标题
使用高斯过程偏好学习和生成图像模型研究面部偏好
DOI：
10.1007/978-3-030-84340-3_15
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Watanabe Katsumi
中科院分区：
文献类型：
--
作者：
Komori Masashi;Shiroshita Keito;Nakagami Masataka;Nakamura Koyo;Kobayashi Maiko;Watanabe Katsumi
这项研究介绍了一种新的方法来研究人类面部吸引力的内在心理物理功能，使用贝叶斯优化（BO）和StyleGAN 2相结合的序贯实验设计。为了从成对比较数据中估计面部吸引力函数，我们使用了结合高斯过程偏好学习（GPPL）的BO。50名日本女大学生提供了面部照片。我们将每个女性面部图像嵌入到在Flickr-Faces-HQ（FFHQ）数据集上训练的StyleGAN 2网络中的潜在表示（维度）中。使用PCA，潜在表征的维度被降低到一个8维子空间，我们在这里称为日本女性人脸空间。9名参与者参与了成对比较任务。他们必须选择在人脸子空间中使用StyleGAN 2合成的更有吸引力的面部图像，并在100次试验中提供他们的评估。前80次试验的刺激是从人脸子空间中随机生成的参数创建的，而剩余的20次试验是从使用采集函数计算的参数创建的。我们估计的面部参数对应的最大，最小，25，50，75百分位数的吸引力等级和重建的脸的基础上的结果。结果表明，StyleGAN 2和GPPL方法的组合是一种有效的方法来阐明人类的复杂刺激，如人脸的kansei评价。
This study introduces a novel approach to investigate human facial attractiveness’s intrinsic psychophysical function using a sequential experimental design with a combination of Bayesian optimization (BO) and StyleGAN2. To estimate a facial attractiveness function from pairwise comparison data, we used a BO that incorporates Gaussian process preference learning (GPPL). Fifty female Japanese university students provided facial photographs. We embedded each female facial image into a latent representation (dimensions) in the StyleGAN2 network trained on the Flickr-Faces-HQ (FFHQ) dataset. Using PCA, the latent representations’ dimension is reduced to an 8-dimensional subspace, which we refer to here as the Japanese female face space. Nine participants participated in the pairwise comparison task. They had to choose the more attractive facial images synthesized using StyleGAN2 in the face subspace and provided their evaluations in 100 trials. The stimuli for the first 80 trials were created from randomly generated parameters in the face subspace, while the remaining 20 trials were created from the parameters calculated using the acquisition function. We estimated the facial parameters corresponding to the most, the least, 25, 50, 75 percentile rank of attractiveness and reconstructed the faces based on the results. The results show that a combination of StyleGAN2 and GPPL methodologies is an effective way to elucidate humankanseievaluations of complex stimuli such as human faces.