dynalytix

What a climbing move felt like, aligned to pose.

Pose estimation can tell you how a climber moved. It cannot tell you what they were trying, how it went, or where it hurt. This project collects that missing layer and aligns it to the pose data, frame by frame.

Open the labeling platform

The Labeling Platform

collect.dynalytix.net · used by pilot labelers

How the data is collected and labeled

Who the Dataset Is For

🧗‍♀️

Climbers

Climbers contribute video and label their own movement. Their account of a move is the part of the data that no camera can supply.

🏥

Physical Therapists

Clinicians who work with climbers have very little movement data tied to reported pain, fatigue, or instability. This dataset is meant to help close that gap.

🔬

Researchers

Researchers in climbing biomechanics get kinematics and structured labels in one table, one row per frame, with a documented schema.

What Is Recorded

Pose data and human labels, joined on every frame

📐

Pose and Joint Angles

33 body landmarks per frame, and 12 joint angles computed from them

⚡

Velocity Vectors

Landmark velocities and center of mass across each move

🎯

Sensation Labels

9 sensation types, tagged by body location (16 sites), side, and a 0 to 10 level, on the frames where they occur

Data Pipeline

Stage	Process	Output
Video Upload	Pose estimation analysis	Joint coordinates & angles
Motion Analysis	Velocity & acceleration calculation	Movement vectors
Labeling	Three lenses per move, sensation tags per frame	One labeled row per frame

See our full methodology

Who Is Behind It, and Where It Stands

People

•Jolie Abadir, project lead and sole developer. Cognitive science at Columbia (Barnard College), competitive boulderer. The project began in 2023 as a climbing tracker she built with OpenCV and MediaPipe, was set down for a stretch, and was picked back up in 2025.
•Taylor Reed. He has advised the project since it started in 2023, rejoined when the work resumed in 2025, and has shaped the labeling schema through multiple rounds of feedback.

Status

•The multi-user labeling platform is live, rebuilt in 2026 on a new schema.
•A pilot with a small group of labelers is beginning. Data collection starts with it, so there is no dataset to cite yet.
•A manuscript is in preparation.
•Videos are stored privately and scoped to each account. A de-identified pose and label dataset is planned for release alongside the paper.
•No model has been trained yet. What the data can support is a question for after the pilot.