SAniHead: Sketching Animal-Like 3D Character Heads Using a View-Surface Collaborative Mesh Generative Network.

Area of Science:

• Computer Graphics
• Artificial Intelligence
• 3D Modeling

Background:

• 3D head modeling is crucial for character design in games and film.
• Existing research primarily focuses on human heads, neglecting diverse and detailed animal head shapes.

Purpose of the Study:

• To present SAniHead, an interactive system for generating 3D animal heads from dual-view sketches.
• To introduce a novel view-surface collaborative mesh generative network for detailed animal head modeling.

Main Methods:

• A graph convolutional neural network (GCNN) deforms a template mesh to fit sketch shapes, creating a coarse model.
• Image-to-image translation networks infer details from vertex maps, which are then refined using a cascaded GCNN and a sketching interface.

Main Results:

• The SAniHead system demonstrates superior performance in generating detailed 3D animal heads.
• Experimental results validate the approach's effectiveness and the interactive system's usability.
• A new dataset of 3D animal heads with line drawings was created.

Conclusions:

• The proposed view-surface collaborative mesh generative network effectively models complex animal head geometry.
• SAniHead offers a practical and efficient solution for creating diverse 3D animal heads for digital applications.