An AI coach that recognizes the dumbbell you pick up, counts your reps and checks your form, running live on a phone with no manual input.
Schematic animation of detection and pose tracking
Logging weights and reps by hand is tedious and error-prone, so the data behind "personalized" training is often wrong or missing. Barcode tags on equipment don't scale across gyms, and every gym's dumbbells look different.
The goal: let the camera do the work. Identify the dumbbell, follow the movement, count every rep and give feedback in real time, on the user's own device.
We gathered workout videos across exercises and dumbbell types. Annotators labeled every dumbbell with its position and type, and the labeled footage became the training set.
A detection model, trained on that dataset and refined across gym settings and equipment types, finds the dumbbell. The system combines detections across consecutive frames to reduce errors from blur and occlusion, and a second model tracks the user's pose for rep counting and form feedback.
We tested on unseen footage and retrained after reviewing the errors, then optimized the models to run on smartphones and wearables, so detection and analysis happen on the device without a cloud connection.
For a project, support with something we built, or anything else, email us and we will get back to you.
Email us