Subdividing a three-dimensional mesh utilizing a neural network

A system comprising:

• a memory component;
• and
• a processing device coupled to the memory component, the processing device to perform operations comprising:
• determining per-vertex feature vectors of existing vertices in a three-dimensional mesh based on a plurality of local geometries comprising adjacent vertices of the existing vertices;
• generating, utilizing a first neural network, a set of displacement coordinates to displace a subset of the existing vertices from initial positions of the subset of the existing vertices to new positions for the subset of the existing vertices by causing the first neural network to generate the set of displacement coordinates from the per-vertex feature vectors of the existing vertices;
• generating, utilizing a second neural network, a set of new coordinates for new vertices along edges connecting the existing vertices by causing the second neural network to generate feature vectors for the new vertices based on the per-vertex feature vectors of the existing vertices;
• generating a subdivided three-dimensional mesh comprising vertices at positions according to the set of displacement coordinates displacing the subset of the existing vertices and the set of new coordinates for the new vertices added to the edges connecting the existing vertices;
• and
• recursively subdividing the three-dimensional mesh to generate a plurality of levels of subdivided three-dimensional meshes according to displacement coordinates for existing vertices generated utilizing the first neural network and new coordinates for new vertices generated utilizing the second neural network.

A non-transitory computer readable medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:

• determining local geometries comprising adjacent vertices of existing vertices in a three-dimensional mesh;
• generating, utilizing a plurality of neural networks, a set of displacement coordinates displacing a subset of the existing vertices from initial positions of the subset of the existing vertices in the three-dimensional mesh to new positions and a set of new coordinates for new vertices along edges connecting the existing vertices by causing the plurality of neural networks to generate the set of displacement coordinates and the set of new coordinates from feature vectors of the local geometries;
• generating a subdivided three-dimensional mesh comprising vertices positioned according to the set of displacement coordinates for the existing vertices and the set of new coordinates for the new vertices added to the edges connecting the existing vertices;
• and
• recursively subdividing the three-dimensional mesh to generate a plurality of levels of subdivided three-dimensional meshes according to the set of displacement coordinates for the existing vertices and the set of new coordinates for the new vertices generated utilizing the plurality of neural networks.