# I made a computer vision tool for gym form analysis — retained record

> Retained reddit record from reddit:computervision.

Canonical: https://highsignal.app/data/records/00b539bfff65e7d5

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:computervision)  published 2026-09-23 13:41 ingested 2026-09-24 02:37

[original source ↗](https://www.reddit.com/r/computervision/comments/1wo6chi/i_made_a_computer_vision_tool_for_gym_form/)

## retained text



```
It runs entirely in the browser with MediaPipe Pose Landmarker at about 30 - 60 fps. Everything above the model is geometry. 

Prop the phone up, pick an exercise, and it turns the skeleton red and beeps when a rule breaks. It also counts reps and sets with a state machine on one trigger metric per exercise.

Exercises are declarative profiles, a trigger metric for reps, rules with min and max, phase gating. Only four so far have been implemented.

Deadlift: torso angle from horizontal (catches hips shooting up), ear-shoulder-hip angle as a spine-neutrality proxy (no mid-spine landmark exists), wrist-to-ankle offset in torso lengths as a bar path proxy.

Lat pulldown: torso lean limit, elbow angle as the rep trigger.

Bench press: forearm tilt from vertical, so wrists stay stacked over elbows.

Biceps curl: upper arm tilt from vertical (detects elbow drift) and torso angle (avoid swinging)

[Here is the link](https://morgan-pt.vercel.app/) if you want to try it out.

Make sure you switch on screen recording if you want to capture your form and range. 

Here is the link to the [GitHub repo](https://github.com/frankndungu/morgan)
```



document fetched 2026-09-24 00:24 · canonical https://reddit.com/r/computervision/comments/1wo6chi/i_made_a_computer_vision_tool_for_gym_form
