spatialhub 0.1.1

A lightweight, PyTorch-free spatial computing and feature matching library.

spatialhub

A lightweight, PyTorch-free computer vision hub for Python.

spatialhub is built for production and fast execution. By removing heavy PyTorch code and using ONNX Runtime, it gives you reliable, server-safe AI models with automatic downloads and zero memory leaks.

Why We Built This

Modern AI research is amazing, but running it in production can be hard due to heavy setups, GPU issues, and high memory usage. spatialhub fixes this by offering:

• No PyTorch Needed: Models are converted to ONNX so they run much faster and are easier to install.
• Auto-Downloading: It automatically downloads model files from Hugging Face so your project folder stays clean.
• Server-Safe: Built with opencv-python-headless so drawing images will never crash your servers or Docker containers.
• Low Memory: It only loads images into memory when you actually ask to draw and save them.

Available Models

Check the specific guides for our current models:

• EfficientLoFTR - Fast and accurate local feature matching.

Roadmap: What's Next

In the future, spatialhub will expand to include more computer vision tools built for real-world use:

• [Planned] Depth Estimation (e.g., DepthAnything v2)
• [Planned] Pose Estimation & Object Tracking
• [Planned] Point Cloud Tools

Installation

Install via pip (requires Python 3.12+):

pip install spatialhub

To enable GPU acceleration, install the optional GPU dependencies:
```bash
pip install "spatialhub[gpu]"

Universal Quickstart

All models in spatialhub share a unified, predictable API structure.

from spatialhub import EfficientLoFTR # Or any future model

def main():
    # Initialize the adapter (weights auto-download on first run)
    model = EfficientLoFTR()
    
    # Run inference on raw paths or numpy arrays
    result = model.match("image1.jpg", "image2.jpg", max_dim=1024)

    # Utilize lazy visualization (safely loads and renders without memory bloat)
    result.visualize(top_k=50, save_path="output.png")

if __name__ == "__main__":
    main()

License

The core spatialhub library is licensed under the Apache 2.0 License.

Note: Individual model architectures and weights downloaded via the library inherit the licenses of their original authors. Please check the specific model documentation directories for details.