Most climbing guide material still lives on paper topos or flat photos of a crag, so an overlay that puts the routes directly onto the rock in AR reads as a genuinely different approach. The maker built this as a masters thesis on visual localization for AR on photogrammetric models, and the demo shows an AR bouldering topo on a camera display while a user walks around an outdoor boulder.
The pipeline pairs visual localization, done with hloc plus XFeat features and nearest neighbor matching, with visual-inertial odometry. An extended Kalman filter fuses the two streams and estimates a similarity transform that maps the VIO frame into the world frame. Synchronization between the two sides is attempted at 1 Hz, and the localization itself carries a 0.5 second delay. Rendering runs on a desktop PC.
Several details stand out. Bright pink lines mark bouldering routes and variations directly on the stone surface, and text labels anchored in 3D show the route names next to each line. The tracking holds those markers and tags to the boulder's faces as the camera moves, and the routes appear in the render once the first visual localization result comes in. The scan and the routes both come from the maker's own 3D climbing topo, which makes the whole thing a tidy closed loop.
The open question is mobile support. The maker says the visual localization should be runnable on a phone, but the thesis scope leaves mobile out, and even if it were added, Android Chrome exposes ARCore while iOS would be out of question for the web app. Whether support arrives depends on feedback after the thesis is submitted.