README.md

metadata

title: DeepFracture Runtime
emoji: 💥
colorFrom: gray
colorTo: red
sdk: static
pinned: false
license: mit
models:
  - nikoloside/deepfracture
tags:
  - fracture
  - physics
  - simulation
  - graphics
  - three-js

DeepFracture — Live Web Runtime

Real-time, in-browser neural brittle-fracture demo: click the object to shoot a projectile; the impact becomes the paper's 7-D collision embedding, a JS mirror of the Siren encoder selects the nearest VQ-VAE codebook entry (nikoloside/deepfracture), and the matching pre-decoded fragment set (decoder + watershed segmentation, baked offline per codebook entry) swaps into the rigid-body simulation.

Three.js + Rapier. Also live at https://nikoloside.graphics/deepfracture-live/.

Paper

This demo accompanies:

DeepFracture: A Generative Approach for Predicting Brittle Fractures with Neural Discrete Representation Learning. Yuhang Huang, Takashi Kanai. Computer Graphics Forum, e70002, 2025. DOI: 10.1111/cgf.70002

@article{huang2025deepfracture,
  author  = {Huang, Yuhang and Kanai, Takashi},
  title   = {DeepFracture: A Generative Approach for Predicting Brittle
             Fractures with Neural Discrete Representation Learning},
  journal = {Computer Graphics Forum},
  pages   = {e70002},
  year    = {2025},
  doi     = {https://doi.org/10.1111/cgf.70002}
}

Papers: DeepFracture (CGF 2025) · Far-From-Boundary Fields (SMI 2026) · Code: TEBP-DeepFracture