Occlusion-robust Stylization for Drawing-based 3D Animation

Occlusion-robust Stylization Framework is consistent 3D animation system from single image which provides occlusion-robust results of body parts under dynamic 3D motions.

3D animation generates videos from an input image and target 3D motion. Drawing-based 3D animation aims to preserve an artist’s unique style, such as rough contours and distinctive strokes. However, current methods struggle under occlusions (e.g., overlapping body parts), causing flickering and style degradation. This is due to a stylization pose gap—stylization networks are trained only on occlusion-free poses, but real inference involves occluded ones. To address this, we propose the Occlusion-robust Stylization Framework (OSF), which uses optical flow to generate occlusion-aware edge guidance for consistent stylization. OSF also improves efficiency, running 2.4× faster and using 2.1× less memory than previous two-stage approaches.