Neural Physics Subspaces
基于 XMAKE + Imgui + OpenGL + pybind11 的神经物理子空间研究项目，复现论文 neural-physics-subspaces
Overview
A neural physics subspaces project that reproduces the paper “Neural Physics Subspaces” using a modern C++/Python hybrid architecture. The system combines physical simulation with learned subspaces for efficient dynamics computation.
Tech Stack
- XMAKE — Build system configuration
- Imgui — GUI and interactive visualization
- OpenGL — 3D rendering engine
- pybind11 — C++ / Python interoperability
- JAX — Automatic differentiation and deep learning
Architecture
The project integrates C++ for performance-critical rendering and simulation with Python for neural network training. The workflow: XMAKE compiles the C++ core → Imgui handles interaction → OpenGL renders the visualization → pybind11 bridges Python training results.
Publications
Related paper: Neural Physics Subspaces (arXiv:2305.03846)
Pre-trained models are available on the GitHub Releases page.