Purifying real images with an attention-guided style transfer network for gaze estimation
Xianping Fu,Yuxiao Yan,Yang Yan,Jinjia Peng,Huibing Wang
DOI: https://doi.org/10.1016/j.engappai.2020.103609
IF: 8
2020-05-01
Engineering Applications of Artificial Intelligence
Abstract:<p>Recently, the progress of learning-by-synthesis has proposed a training model for synthetic images, which can effectively reduce the cost of human and material resources. Image synthesis has been widely accepted as a cost effective way to learn models because it provides training sets that are large, diverse and accurately labeled. However, the realism of the synthetic image is not enough, this affects generalization on naturalistic test image. In an attempt to address this issue, previous methods learn a model to improve the realism of synthetic image. Different from previous methods, we take the first step towards purifying the real image to weaken the influence of light and convert the distribution of an outdoor naturalistic image through a real-time style transfer task to that of indoor synthetic image. In this paper, we first introduce the segmentation masks to construct Red, Green, and Blue-mask (RGB-mask) pairs as inputs, then we design an attention-guided style transfer network to learn style features separately from the attention and background regions, learn content features from full and attention regions. Moreover, we propose a novel region-level task-guided loss to restrain the features learnt from style and content. Experiments were performed using a mixed research (qualitative and quantitative) method to demonstrate the possibility of purifying real images in complex directions. We evaluate the proposed method on three public datasets, including Labeled pupils in the wild (LPW), Common Objects in COntext (COCO) and MPIIGaze. Extensive experimental results show that the proposed method is effective and achieves the state-of-the-art results.</p>
automation & control systems,computer science, artificial intelligence,engineering, electrical & electronic, multidisciplinary