What a climbing move felt like, aligned to pose.
Dynalytix is a research project. We are building a labeled dataset of climbing movement, and the platform used to collect it, so that pose data can be studied alongside what the climber was trying and what it felt like. Led by Jolie Abadir with Taylor Reed.
Get updatesThe main project · active
Climbers upload video and describe their own movement. Each move is labeled through three lenses (what the wall offered, what the climber chose, how it turned out), and sensations like pain, instability, or fatigue are tagged on the exact frames where they happened. Every label is joined to 33 pose landmarks, 12 joint angles, velocities, and center of mass for that frame.
Live
Multi-user labeling platform
Pilot beginning
A small group of labelers
Manuscript in preparation
With Taylor Reed
Also here · paused
Movement scoring prototype. An earlier part of this project: rule-based deep squat scoring from front and side video, against published thresholds.
See the prototype →Video of a boulder or route attempt is uploaded to the labeling platform. Videos are stored privately, scoped to the uploader's account, and used only for labeling.
MediaPipe finds 33 body landmarks in every frame. Joint angles, velocities, and center of mass are computed from them.
Labelers describe each move through three lenses (environment, strategy, outcome) and tag sensations frame by frame: type, body location, side, and a 0 to 10 level.
The labeling platform is live at collect.dynalytix.net, and a pilot with a small group of labelers is beginning.
A manuscript is in preparation. Data collection starts with the pilot, and a de-identified pose and label dataset is planned for release alongside the paper.
The code is on GitHub: Jolieabadir/dynalytix
Get occasional updates on the dataset and the paper, or tell us you would like to help label climbing video in the pilot.