A model training method, device, equipment and medium based on skin weight

Abstract

The embodiment of the present application discloses a model training method, device, device, and medium based on skin weights. The method includes: obtaining a sample wearable item model with a sample weight label, extracting a sample bone structure; based on the vertex position information of the model vertex and the normal vector information to determine the basic characteristics of the sample vertices; determine the sample vertex connection relationship matrix based on the vertex topology; voxelize the sample wearing item model to obtain the voxel of the sample wearing item model, and determine the geodesic distance feature of the sample based on the voxel; The sample vertex basic features, sample vertex connection relationship matrix, sample geodesic distance feature and initial network model are used to obtain the predicted weight label. Based on the predicted weight label and sample weight label, the initial network model is trained to obtain the target skin weight. target network model. By adopting the embodiments of the present application, the accuracy of predicting skin weights can be improved.

Description

technical field [0001] The present application relates to the field of computer technology, and in particular to a skin weight-based model training method, device, equipment and medium. Background technique [0002] At present, in the animation production process, since the wearable object model (for example, a 3D human clothing model) and the bone structure distributed in the 3D human clothing are independent of each other, in order to let the bone structure drive the 3D human clothing model to produce reasonable motion In the process of creating the 3D human clothing model, the computer device with skinning function needs to set the skinning weight between the vertices of the model and the bone structure to achieve the effect of binding the 3D human clothing model to the bone structure. [0003] It can be understood that the existing skinning algorithm can determine the distance feature corresponding to the model vertex and the bone structure by calculating the Euclidean d...

AI Technical Summary

Problems solved by technology

However, in practical applications involving curved surfaces, the Euclidean distance cannot accurately represent the distance between the bone structure and the model vertices on the 3D human clothing model, which leads to inaccurate final skin weight settings

For example, in the case of a three-dimensional human clothing model standing naturally, a certain bone structure 1 (for example, the bone structure where the elbow joint is located) may be placed close to the ribs, so the bone structure 1 determined by the computer device and the model of the area under the ribs The Euclidean distance between the vertices is small, so that the skin weight between the finally determined model vertices of the subcostal area and the bone structure 1 is relatively large, that is, the model vertices of the subcostal area are largely influenced by the elbow joint. In fact, the vertices of the model in the subcostal area should be less or even not affected by the elbow joint. In this way, when the 3D human clothing model moves with the bone structure, the 3D human clothing model may have severe deformation such as the face , interspersed and other abnormal conditions, which reduces the accuracy of the skin weight

Images