“Train Once, Generate Anywhere: Discretization Agnostic Neural Cellular Automata Using SPH Method” by Kim and Park
Conference:
Type(s):
Title:
- Train Once, Generate Anywhere: Discretization Agnostic Neural Cellular Automata Using SPH Method
Session/Category Title:
- Images, Video & Computer Vision
Presenter(s)/Author(s):
Abstract:
We introduce SPH‑NCA, a discretization agnostic neural cellular automata that uses a differentiable SPH method for perception and a stable training scheme, allowing image and texture synthesis on any grid, resolution, or 3D surface while trained on a fixed-resolution 2D image.
References:
[1] Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuhmann, Ludwig Schmidt, and Jenia Jitsev. 2023. Reproducible scaling laws for contrastive language-image learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2818–2829.
[2] Daniele Grattarola, Lorenzo Livi, and Cesare Alippi. 2021. Learning graph cellular automata. Advances in Neural Information Processing Systems 34 (2021), 20983–20994.
[3] Alexander Mordvintsev and Eyvind Niklasson. 2021. μ NCA: Texture Generation with Ultra-Compact Neural Cellular Automata. arXiv preprint arXiv:https://arXiv.org/abs/2111.13545 (2021).
[4] Alexander Mordvintsev, Ettore Randazzo, Eyvind Niklasson, and Michael Levin. 2020. Growing neural cellular automata. Distill 5, 2 (2020), e23.
[5] Ehsan Pajouheshgar, Yitao Xu, Alexander Mordvintsev, Eyvind Niklasson, Tong Zhang, and Sabine Süsstrunk. 2024. Mesh neural cellular automata. ACM Transactions on Graphics (TOG) 43, 4 (2024), 1–16.
