[Feature Request] Add XFeat Learned Keypoint Detector & Descriptor

[tak2hu]

XFeat: Accelerated Features for Lightweight Image Matching TL;DR: Really fast learned keypoint detector and descriptor. Supports sparse and semi-dense matching.

Deep learning based feature detectors are usually more robust but are heavy using CPU. I got 1 FPS using the SuperPoint demo but 5 FPS with the XFeat demo (not much ik but still) using an Intel Core i7 Gen 8 laptop. It's a nice alternative to SuperPoint and I'm curious how well it would integrate into RTAB-Map.

Also with LightGlue learned matcher would be nice too.

[matlabbe]

With right python interface, they could be probably be used in rtabmap without changing c++ code. See examples: https://github.com/introlab/rtabmap/tree/master/corelib/src/python

Related posts:

• https://github.com/introlab/rtabmap/issues/1129
• https://github.com/introlab/rtabmap/issues/1123
• https://github.com/introlab/rtabmap/issues/1148