Interactive demo of the paper “DeepFracture: A Generative Approach for Predicting Brittle Fractures with Neural Discrete Representation Learning” — Yuhang Huang & Takashi Kanai, Computer Graphics Forum, 2025.
Click the squirrel to shoot. Drag to orbit, scroll to zoom.
Impact → 7-D collision embedding → Siren encoder + VQ codebook lookup (live in JS) → fragments re-enter the physics. Every codebook entry was decoded (128³) and watershed-segmented offline, so the runtime result is identical to the paper's pipeline with zero inference wait.