A progressive self-distillation emotional state estimation method for viewpoint and environment interaction

Abstract

The application discloses a kind of view point and environment interaction's progressive self-distillation emotional state estimation method, to realize the accurate determination of emotional state by collecting the eyeball view point data of viewer under different images.First, use high-precision eye tracking equipment to form initial data set, and the image is segmented into multiple view point stay area, according to the stay order to carry out segmentation and sequencing.Then, use preliminary deep learning model to carry out difficulty evaluation to image sample, to support progressive learning.The method designs deep neural network, and combines self-distillation mechanism to optimize prediction result constantly.Model learns gradually from easy to difficult, through high confidence prediction to guide current stage learning, finally assesses effect on independent verification set.The method effectively improves the accuracy and robustness of emotional state estimation, applicable to psychology research, intelligent human-computer interaction and user experience optimization and multiple fields, with high precision and multi-scene application ability, provides reliable emotional state analysis result.

Images