Parameter estimation model training method and device, equipment and storage medium

Abstract

The invention discloses a parameter estimation model training method and device, equipment and a storage medium. The method comprises the following steps: inputting a face image training set into a pre-constructed neural network model, estimating reconstruction parameters specified by three-dimensional face reconstruction, inputting the reconstruction parameters into a pre-constructed three-dimensional deformation model, and reconstructing a three-dimensional face corresponding to a training sample; calculating loss functions between the three-dimensional face and the training sample under multiple items of two-dimensional supervision information, and adjusting the weight corresponding to each loss function; and generating a corresponding fitting loss function based on each loss function and the corresponding weight, and performing reverse correction on the neural network model by using the fitting loss function to obtain a trained parameter estimation model. According to the technicalscheme provided by the invention, the parameter estimation model is trained through the loss function under multiple items of two-dimensional supervision information, so that the reference information in the training process is more comprehensive, and the estimation accuracy of the reconstruction parameters adopted during three-dimensional face reconstruction is improved.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of image processing, and in particular, to a training method, device, device and storage medium for a parameter estimation model. Background technique [0002] With the development of video technology, there is an increasing demand for creating realistic face models in entertainment applications such as face animation, face recognition, and augmented reality (Augmented Reality, AR), which require face display. At present, it is very difficult to create a realistic 3D face model. It is necessary to reconstruct the corresponding 3D face from one or more 2D face images or depth images, so that it includes face shape, color, illumination and head. Various types of 3D information such as the internal rotation angle. [0003] In the existing 3D face reconstruction methods, a large amount of 3D scan data of faces is usually collected first, and then the 3D scan data is used to construct a cor...

AI Technical Summary

Problems solved by technology

[0004] At present, the estimation of the reconstruction parameters corresponding to each principal component base is usually to directly use the pixel value of each feature point in the 2D face image as the supervision information of the 3D face deformation to estimate the reconstruction parameters corresponding to each principal component base. However, since it is a pathological problem from 2D images to 3D reconstruction, only the pixel values ​​of feature points are used as supervision information, which cannot guarantee the estimation accuracy of reconstruction parameters; or, use the current multi-view 2D face image or Depth information is used as input to estimate the reconstruction parameters corresponding to various principal component bases, but it needs to collect multiple face images and even special sensors to collect depth images, resulting in limited reconstruction scenes, and too many requirements for the collection of reconstruction parameters, making the 3D human The operation of face reconstruction is more cumbersome and complicated

Images