Deep learning fixation prediction method and system based on face appearance

The invention discloses a deep learning fixation prediction method and system based on face appearance. The method comprises the following steps: acquiring a face appearance picture of a to-be-detected object; processing the face appearance picture of the to-be-detected object, and extracting a face picture; and inputting the face picture into a trained fixation prediction model, and predicting a human eye fixation direction in the face picture. According to the method, the convolutional neural network, the window multi-head attention mechanism and the moving window multi-head attention mechanism are combined, so that the image local spatial feature learning ability and the global feature modeling ability of the network are improved; the problems that an existing gaze prediction method based on a visual transformer cannot achieve multi-scale feature learning and image global self-attention calculation is difficult are solved. 本发明公开了一种基于人脸外观的深度学习注视预测方法及系统，方法包括：获取待测对象的人脸外观图片；对所述待测对象的人脸外观图片进行处理，提取人脸图片；将所述人脸图片输入训