Results our clients ship with

The challenge

Smartan, an AI-powered fitness technology company, needed large volumes of precisely annotated human movement data to train their exercise-form analysis models. Workout footage is unforgiving to annotate — fast motion, occluded joints, unusual body positions, and varied camera angles — and inconsistent keypoints would translate directly into wrong form-feedback for end users.

What we did

Vidyut Data assembled a dedicated annotation team trained on anatomical keypoint conventions. We annotated skeletal keypoints frame-by-frame across workout videos — joints, limb orientation, and visibility flags — with left/right symmetry validation and temporal-consistency checks so keypoints never jittered between frames. Edge cases like occluded joints during push-ups or overlapping limbs in compound movements were escalated, decided, and documented into a living labeling guide.

“The movement annotation quality directly improved our pose model's feedback accuracy. The team understood anatomy, not just boxes and points.”