LightGlue-ONNX: High-Speed Neural Feature Matching
Primary Source: https://github.com/fabio-sim/LightGlue-ONNX
Author: Fabio Sim · Year: 2026 · Publisher: Fabio Sim
Summary
An optimized ONNX implementation of LightGlue enabling real-time, lightning-fast local feature matching and camera relative pose estimation across TensorRT, OpenVINO, and web runtimes.
Key Points
- Adaptive deep neural network that adjusts computation time based on image matching difficulty.
- Cross-platform deployment with zero PyTorch runtime dependency.
- Essential building block for robotics, visual SLAM, and augmented reality tracking.