Human eye image processing model training method, image processing method and device

Abstract

The present invention provides a training method for a human eye image processing model, comprising: processing different human eye images to be corrected in the human eye image set to be corrected through the style transfer network of the human eye image processing model to form Target human eye image; determine the training samples that match the usage environment of the human eye image processing model, and train the correction network of the human eye image processing model through the training sample set that matches the usage environment of the human eye image processing model, Model parameters adapted to the correction network of the human eye image processing model are determined. The invention also provides an image processing method, device, electronic equipment and storage medium. The invention improves the training efficiency of the human eye image processing model while ensuring the recognition accuracy of the human eye image processing model, reduces the complexity of human eye image processing, adapts to different data processing environments, and enhances the robustness of the human eye image processing model , to reduce the cost of use.

Description

technical field [0001] The invention relates to image processing technology, in particular to a training method of a human eye image processing model, an image processing method, a device, electronic equipment and a storage medium. Background technique [0002] In the process of video processing, Gaze Correction is to process the pictures of human eyes, which can change the direction of sight of human eyes in the video. Eyesight correction has practical value and broad prospects in communication scenarios such as video calls. However, since the image or video of the human eye will have large changes in size, resolution, viewing angle, illumination, texture, occlusion, etc., the image rendered based on 3D modeling is usually different from real people in terms of texture and material. There is a large gap in eye images, lack of naturalness, and the hardware cost of using 3D equipment to achieve line of sight correction is relatively high. However, due to the lack of large-sc...

AI Technical Summary

Problems solved by technology

However, since the image or video of the human eye will have large changes in size, resolution, viewing angle, illumination, texture, occlusion, etc., the image rendered based on 3D modeling is usually different from real people in terms of texture and material. There is a large gap in eye images, lack of naturalness, and the hardware cost of using 3D equipment to achieve line of sight correction is relatively high. However, due to the lack of large-scale training samples, it is necessary to collect and label samples to achieve line of sight correction through deep neural networks, which increases the use of cost

Images