Methodology

We break your footage down step by step: video → pixels → joints (head, knees, elbows and more) → deep learning position classification, with a confidence score → analytics built around you.

97% of frames labeled correctly on footage the model never saw in training.

Getting the best results

• Keep the camera still, with both athletes fully in frame.
• Film from the edge of the mat, slightly elevated if you can.
• Record one roll per video, with as few other people in frame as possible.
• Tell us which athlete is you when you upload.