Inquiry about Commercial Use

[ZHR-rgb]

Hello,

I’ve tried different feature extraction networks, such as DISK, Dedode, XFeat, and ALIKED, and I found that SuperPoint works best for detecting real keypoints. However, I’m not sure about the commercial use of your implementation, since the official SuperPoint license doesn’t allow that.

I would appreciate any guidance you can provide on this matter.

Thank you very much!

Best regards

[rpautrat]

Hi, I agree with you that SuperPoint delivers the best "corner-like" keypoints among the ones you cited.

This repo is a reimplementation from scratch of the original SuperPoint, solely based on the original paper. So the license of this work is open and allows for commercial applications in theory. I don't know if Magic Leap did a patent for this invention though.

[ZHR-rgb]

Thank you for your patience and thoughtful responses!
I feel that SuperPoint clearly defines what constitutes a keypoint during training, using supervised learning on synthetic data and self-supervised learning on real-world images. Is this training strategy key to what makes SuperPoint so effective?