Why this matters
Bridging model development and visualization is a persistent friction point for animation research: models are trained in Python but validated in game engines, creating cumbersome export/streaming steps. This project removes that gap by running training, inference, and real-time rendering together on NumPy/PyTorch, letting researchers iterate on motion models and immediately inspect results — even backpropagating through inference when needed.
What Sets It Apart
- Unified Python workflow: Training, inference, feature extraction, and rendering share the same backend (NumPy / PyTorch). So what: you can instrument models and visualizations during a single run without ONNX or streaming bridges.
- Game-engine-style architecture with ECS & update loops: So what: it preserves familiar lifecycle semantics for animation engineers, making it easier to port ideas from Unity-style pipelines while staying in Python.
- Motion import & format support: GLB/FBX/BVH → internal .npz format with 3D positions and quaternions. So what: common mocap datasets plug in directly for training and evaluation.
- Real-time renderer + headless/standalone/manual modes: So what: supports interactive demos locally and headless training on servers, enabling both rapid prototyping and scalable experiments.
- Research-first features: supports MLPs, autoencoders, codebook matching, quantization-friendly PyTorch models, and permits backprop through inference. So what: enables experimental workflows not possible when visualization is decoupled.
Who It's For & Trade-offs
Great fit if you are an animation researcher or engineer who needs rapid iteration between model changes and visual results, or if you want a Python-native alternative to Unity-based pipelines for mocap-driven controllers and motion synthesis. It’s also useful for building interactive demos and forrerunning experiments without engine interop.
Look elsewhere if you need a production-ready commercial game engine or enterprise middleware: rendering and tooling are research-oriented, the license is CC BY‑NC 4.0 (non-commercial), and some features—physics, path planning, advanced audio—are listed as forthcoming. For high-performance, platform-integrated runtime in commercial games, an engine like Unity/Unreal remains the pragmatic choice.
Where it fits
This project occupies the space between model-research toolkits and game-engine runtimes: it reduces iteration cost for animation model development and is most valuable when the goal is model exploration, visualization-driven debugging, or building research demos within a Python-first stack.
