Method and device for training depth prediction model and generating face depth image

Abstract

The embodiment of the present invention relates to the field of image processing, and discloses a method and device for training a depth prediction model and generating a depth image of a face, and a method for training a depth prediction model, including: inputting a face image into a preset depth prediction model, Obtain a depth image and a face mask, the depth prediction model at least includes a face depth prediction network branch and a face mask prediction network branch; determine a loss value according to the depth image and the face mask; according to the The loss value adjusts the training parameters of the depth prediction model. It can provide strong constraints on the boundary between the face edge and the background, so that it can effectively alleviate the continuous inaccuracy of the depth value near the face contour in the depth image and the sticking of the depth value of the face part and the background part.

Description

technical field [0001] Embodiments of the present invention relate to the field of image processing, and in particular to a method and device for training a depth prediction model and generating a depth image of a human face. Background technique [0002] In application scenarios such as face-swiping payment, virtual reality / augmented reality (Virtual Reality / Augmented Reality, VR / AR), it is necessary to perform 3D reconstruction or face recognition based on the collected face images. In the process of 3D reconstruction or 3D face recognition, how to obtain face depth information is particularly important. With the development of neural network technology, more and more considerations are given to training the depth prediction model of face images based on face images, so that the trained depth prediction model can be used to process the input face images and output the face image The depth prediction result, that is, the depth image, and then perform 3D face reconstruction...

AI Technical Summary

Problems solved by technology

[0003] However, the accuracy of the depth prediction model trained by the above method is not high, and there are problems in the output face depth image that the depth values ​​near the contour of the face are continuously inaccurate and the depth values ​​​​of the face part and the background part are glued together.

Images