LightGlue

LightGlue is a state-of-the-art, deep neural network for adaptive local feature matching: it delivers high accuracy at light speed.

LightGlue is a high-performance, deep neural network for sparse local feature matching, developed by researchers from ETH Zurich and Microsoft Mixed Reality & AI Lab (presented at ICCV 2023). It significantly improves upon its predecessor, SuperGlue, by incorporating adaptive mechanisms: the network dynamically adjusts its depth and prunes unmatchable key points early. This efficiency gain is substantial: LightGlue achieves a 4-10x speedup over SuperGlue while maintaining comparable or superior accuracy. The model is competitive with slower dense matchers (like LoFTR) but operates at a fraction of the inference time, making it ideal for latency-sensitive applications such as real-time 3D reconstruction and visual localization.

https://github.com/cvg/LightGlue