Rep counter: how a phone camera counts valid reps and calls no-reps.

How does an app count reps?

A rep counter does not watch the whole body and decide that a repetition happened. It follows one landmark per station — the hips for a burpee, the wrists for a snatch, the shoulders for a pull-up — and counts complete cycles of that one signal.

Each set gets two thresholds of its own, placed 28% to 30% of the set’s own range in from the top and the bottom of the movement. A cycle counts only when the signal leaves its resting position, crosses both thresholds and comes back. A half rep that turns around between them, or a twitch at the top, never crosses both, so it is not counted at all — not even as a no-rep.

Two more limits stop noise from becoming reps. The signal has to travel a minimum distance within the cycle, 1.5% to 10% of the frame depending on the station, and a new rep cannot complete sooner than the station’s minimum rep time after the last one. The thresholds adapt to the athlete’s own range; the minimum travel does not, which is why a phone placed too far away can miss reps.

• How a camera frame becomes pose landmarks in the first place — computer vision exercise tracking.

Valid rep or no-rep: what happens after a cycle is counted

A counted cycle is not yet a valid rep. 40 of the 41 stations declare standards that every rep has to show — squat depth, a full overhead position, elbow lockout, a full hang, feet off the floor — and each cycle is checked against them using only the frames inside that cycle. A clearly observed miss turns the cycle into a no-rep, with the reason attached. A rep that takes longer than the station allows, or travels far less or far more than the movement plausibly can, is a no-rep too.

Evidence the camera could not take is treated differently from evidence that a rep failed. If the relevant joints are poorly visible, or the reading sits too close to the threshold to call, the rep is counted and marked as uncertain rather than thrown away. Joint angles are only judged from a side or oblique view in the browser counter: a knee or elbow angle seen head-on is foreshortened, so from the front it is reported as not verifiable instead of as a fault.

• The depth, lockout and extension thresholds used to score technique, movement by movement — the movement standards.

Rep counter rules for all 41 stations

Travel is measured as a share of the video frame, not in centimetres, and angles are measured at the joint: 180° is a fully straight arm or leg, so a smaller knee angle is a deeper position. Select a station for its full movement page where one exists.

Why a rep didn’t count

Almost every missing rep has one of four causes, and each is tied to a specific condition in the station’s config. A whole set rejected for the same reason is far more often the camera than the athlete, so check the framing before changing the movement.

• The identity check failed. Before counting, the whole clip has to look like the selected station: arms overhead in a station that forbids it, a deep squat where none is allowed, travelling across the floor in an in-place station, or a required position never seen at all. A clip that fails gets no count at all, rather than a count against the wrong rules.
• The movement was too small. The whole clip has to move the station’s main landmark by 2% to 12% of the frame, and each cycle has to travel the station’s minimum. A camera too far away makes every movement small, and a partial cycle is never counted.
• The camera view was not accepted. Each station lists the views it trusts, and the most common view in the clip has to be one of them. Only 14 of the 41 stations accept a front-on camera; for the rest, film from the side or at 45°.
• Visibility was too low. The athlete’s landmarks have to be visible across the clip to the station’s level, 35% to 40% depending on the station. Backlighting, loose clothing, equipment in front of a joint or a crop at the feet all lower it.
• Fix the framing first — the camera position guide.

What the rep counter cannot judge

It is not a competition judge. An event standard can require things a camera cannot see from one phone: a ball touching a target, a chest touching the floor, a bar touched with the hips, a precise box height. The counter checks the body positions it can measure and leaves the rest to a human.

Two-dimensional pose estimation measures visible joint positions. It does not measure force, load or medical safety, and any count is only as good as the clip it came from. The honest test is a set you have already counted yourself: run it through the demo and compare.

• Burpee rep counter — what counts as a complete burpee, and pace across a long set.
• Create a free account to save counted sets and see which reps were no-reps, and why.