Marathon road sample

Kelvin Kiptum

Kiptum's clip moves the system away from track footage. The pose read is harder, but that makes it useful for testing road angles and pacing groups.

Road-race rhythm, higher hip-motion proxy, and lower pose visibility because the camera angle is less controlled.

Stride readout

These are clip-derived signals, not a coaching diagnosis. The point is to show what the pipeline can read and where the video still needs review.

Knee sweep

124 deg

Average left/right knee angle range in the processed clip.

Hip bounce

15.9%

Normalized hip vertical oscillation proxy from pose landmarks.

Readable frames

96%

Frames where the pipeline kept enough pose signal to use the stride.

3.56s103 of 107 usable frames

A road clip with a pacing pack and a less stable broadcast angle.

Stride rhythm proxy

1.16

Peak spacing in the lower-body motion signal.

Knee lift proxy

79%

Relative vertical knee travel in the normalized pose.

Pose visibility

46%

Mean landmark visibility across the usable clip.

The camera angle lowers pose visibility, especially compared with the track samples.
The hip-motion proxy is high, so the page flags this as a clip to review rather than a clean comparison target.
This is the best current sample for testing whether the pipeline can leave the track setting.

Runner isolation

The selected runner mask over the source segment.

Stride skeleton

Pose landmarks rendered frame by frame.