GlueStick

Why we included this project

Most image matchers work on points alone, but GlueStick matches points and line segments together in a single deep network. That joint approach helps when you need to align two images of the same scene for structure-from-motion, visual localization, or 3D reconstruction, because lines carry structural cues that survive drastic viewpoint and illumination changes where point-only matchers often fail. The result is more robust matching on low-texture or repetitive scenes. The repository ships pretrained weights and a straightforward inference command, and a Colab notebook lets you try it without installing anything locally. The SuperPoint backbone carries a non-commercial license, so teams with commercial use in mind should factor that in before adopting it. Training code lives in the companion GlueFactory framework, which also covers other matchers like LightGlue.