25-04-2023 дата публикации
Номер: CN116012903A
Принадлежит:
The invention provides a face expression automatic labeling method and system, and the method comprises the steps: obtaining a face image, and selecting a face image with a peak expression as an expression image data set; selecting a one-fifth face image in the data set, and manually marking emotion labels of expressions of the one-fifth face image; and constructing a model framework based on self-supervised learning. In the auxiliary task, an Officient-CapsNet model is used as an encoder for facial expression feature extraction, unlabeled expression images in the expression image data set are sent into the encoder for feature extraction, feature representation is compared and learned, and a pre-training model is obtained; in downstream tasks of self-supervised learning, training and fine tuning are carried out on a labeled data set obtained by the pre-training model, and a final automatic labeling model is obtained; and automatically labeling the facial expression images obtained in the ...
Подробнее