Metahuman Synthetic Dataset for Optimizing 3D Model Skinning

In this study, we present a method for creating a synthetic dataset using the MetaHuman framework to optimize the skinning of 3D models. This study focuses on improving the quality of skeletal deformation (skinning) through leveraging a diverse array of high-fidelity virtual human models. Using MetaHuman, we generated an extensive dataset made up of dozens of virtual characters with varied anthropometric features and precisely defined skinning weight parameters. These data were used to train an algorithm that optimizes the distribution of skinning weights between bones and the model surface. The proposed approach automates the weight rigging process, significantly reducing manual effort for riggers and increasing the accuracy of deformations in animation. Experimental results show that leveraging synthetic data reduces skinning errors and produces smoother character movements compared to traditional methods. The outcomes have direct applications in the video game, animation, virtual reality, and simulation industries, where the rapid and high-quality rigging of numerous characters is required. The method can be integrated into existing graphics engines and development pipelines as a plugin or tool, facilitating the adoption of this technology in practical projects.