htoo.dev

Session: 21-01-2026 .parq

Been having a lot of fun with the integration of my climbing hold detection ML model into the app.

I started to train the machine learning model with a combination of custom and public climbing holds dataset. It’s based on YOLO26; exported to Swift with coreML. And the route prediction is computed with heuristics.

It’s still a lot left to do as u may see, there’re some false positives. The main issue lies with the color filtering as it’s very difficult with the holds that are covered in chalks.

All and all, I’m just excited and having fun building everything from scratch.

Thought I’d share a progress on something I’ve been working on. It’s a grading/reviewing tool for bouldering movements. Input is solely vision based and all processings are done on device and in realtime.