You point your phone at a bouldering wall, snap a bunch of photos, and sometime later a fully navigable 3D model of the wall shows up in your browser. Routes highlighted, holds detected, the whole thing spinnable and zoomable like a little diorama of your gym. That's Climbuddy, and the demo is worth five minutes of your day.

The project comes from Tomáš Sláma and Matěj Kripner, two Czech computer science grads who apparently couldn't decide between photogrammetry and bouldering, so they chose both. The pipeline stitches together COLMAP for structure-from-motion, multi-view stereo texturing, LaMa for inpainting gaps, and then YOLO paired with SAM to detect and segment individual holds. About 100 photos of 20 boulders, 20 minutes of compute, and out comes a textured mesh with routes already identified. The WebGL viewer on top ties it all together: click a route, see the holds light up, log your sends, argue about grades in the comments. It's a lot of moving parts, and the fact that they converge into something this coherent is quietly impressive.

Go poke at the demo, rotate the model, click some routes. It's the rare project where the computer vision, the 3D rendering, and the actual use case all make sense together.