🧬 Cells2Pixels: Scalable Neural Cellular Automata
The Cells2Pixels project has been introduced, allowing for the generation of high-resolution textures and complex biological structures using neural cellular automata (NCA). Instead of working with massive grids, the model uses a sparse grid combined with an LPPN decoder to render detail at arbitrary resolutions.
🌍 It solves the problem of quadratic growth in computational costs in NCA. This hybrid approach paves the way for creating procedural high-quality textures and dynamic objects suitable for real-time rendering.
👤 It is now possible to create detailed self-organizing structures for 3D models without overloading the GPU with pixel-level computations.
Source 1: https://cells2pixels.github.io/ Source 2: https://github.com/TheDevilWillBeBee/Cells2Pixels