movementaIn dev
Peronace Performance Coaching

Climb Analysis

We adaptively sample keyframes across the full video — denser around dynos, foot cuts, and rests — plus a high-resolution reference frame, then run AI movement analysis.

Tip (per JMIR mhealth e82412): place the camera ~3–4 m from the wall at hip height, perpendicular to the climb plane, for best pose estimation accuracy.

Leave blank if the video was filmed today. Setting this when uploading older footage keeps trend graphs and analyses on the right calendar position.

Used to filter dashboard trends — separate competition performance from day-to-day training.

Telling us the outcome anchors the top-out inference and keeps feedback coherent. On a rope, "reached the end with falls / rests" counts as reaching the anchor but not as a clean send; "recording cut" tells the analyzer not to treat the abrupt ending as a fall.

If the route is set in one color (common in gyms), tell us which — the AI will only consider on-color holds as on-route and won't suggest using off-color holds in its feedback.

If multiple people appear (spotter, belayer, another climber), this helps the AI analyse the right person. We'll also ask you mid-flow if our pose tracker spots more than one.

If auto-tracking still struggles, cropping the video before upload is the strongest fallback.

On-device pose tracking
MediaPipe Pose runs over the whole clip in your browser. Tracks the climber across multiple people via persistence + upward motion.
Robust mode (3 local pose runs)
Runs MediaPipe pose tracking 3 times with different sampling rates and reports each numeric movement, joint and load score as mean ± SD. Fully local — no extra AI tokens. Adds ~2× pose-tracking time.
Analysis density
Frame budget scales with clip duration and motion. Defaults keep short and long clips at the same per-second granularity.

Pick a video file to enable Run analysis.

Sessions

Latest analysis pipeline: 2026.07.19

Loading sessions…