基于生成对抗网络的注视估计方法

Gaze Estimation Method Based on Generative Adversarial Networks

基于深度学习的注视视估计研究由于数据收集成本过高,数据标注难度过大,导致很难训练出足够准确且健壮的模型。为减少模型对原始数据的依赖并提高注视估计的准确度,根据轻量级网络构建新的生成对抗网络模型(Light Deep Convolution Generative Adversarial Network,LDCGAN),并构建了新的损失函数。为证明LDCGAN网络的可行性,在实验中与其他注视估计方法进行比较,结果表明LDCGAN模型的预测误差低于其他方法。

Deep learing-based gaze estimation research is difficult to train accurate and robust models due to the high cost of data collection and the difficulty of data annotation.In order to reduce the dependence of the model on the original data and improve the accuracy of fixation estimation,a new Generative Adversarial Network model Light Deep Convolution Generative Adversarial Network(LDCGAN)was constructed based on the lightweight network,and a new loss function was constructed.In order to prove the feasibility of LDCGAN network,the experimental results show that the prediction error of LDCGAN model is lower than that of other fixation estimation methods.