Skip to content

README.md

Latest commit

 

History

History
120 lines (86 loc) · 4.29 KB

README.md

File metadata and controls

120 lines (86 loc) · 4.29 KB
Raw

FormFit

FormFit is a watch-first rehabilitation and form-feedback platform built around Apple Watch motion capture, an iPhone companion app, and a lightweight machine learning pipeline for movement-quality scoring.

This repository has been updated from the original watch-only collector into a fuller Spring 2026 snapshot of the project: branded phone and watch UI, end-to-end watch-to-phone session transfer, app-side session storage, and the PyTorch-to-Core ML model pipeline used to prepare future in-app scoring.

Highlights

  • End-to-end Apple Watch motion collection using CMDeviceMotion
  • iPhone companion app for receiving transferred sessions and storing them in-app
  • Branded phone and watch UI with synchronized visual language and app icons
  • Data pipeline for collecting, labeling, splitting, and training movement data
  • 1D CNN training workflow with Core ML export included in the repo
  • EOH 2026: FormFit earned 2nd Place in Most Industry Impact after presenting to 30,000+ attendees

System Overview

FormFit currently spans three connected layers:

  1. Watch collection The Apple Watch records user acceleration, gyroscope readings, and roll/pitch/yaw orientation data during rehab or strength movements.

  2. Phone experience The iPhone app receives session files from the watch, preserves the raw CSV data, and imports sessions into app-owned storage so the data is ready for analysis and future scoring flows.

  3. ML pipeline The ml/ section contains the training code, labels, split rep files, trained weights, training curve, and exported Core ML package used to support movement-quality scoring.

Repository Structure

.
├── ml/
│   ├── CNN.py
│   ├── FormFitModel.mlpackage/
│   ├── best_model.pth
│   ├── formfit-data/
│   ├── formfit-labels.csv
│   ├── training_history.png
│   ├── README.md
│   └── requirements.txt
├── tools/
│   └── generate_formfit_icons.swift
└── watchos/
    ├── FormFitWatch/
    │   ├── FormFitPhone/
    │   ├── FormFitWatch Watch App/
    │   └── FormFitWatch.xcodeproj
    └── README.md

What Is Included

Apple Watch + iPhone app

  • Watch motion capture flow with start/stop/save controls
  • WatchConnectivity transfer from watch to phone
  • iPhone-side import and persistence of workout sessions
  • UI work for home, workout, progress, profile, tutorial, and leaderboard screens
  • Themed watch and phone styling plus branded FormFit app icons
  • Foundations for future coaching and scoring workflows

Machine learning workflow

  • Rep-window training data under ml/formfit-data/
  • Label file under ml/formfit-labels.csv
  • PyTorch CNN training and scoring script in ml/CNN.py
  • Trained checkpoint in ml/best_model.pth
  • Exported Core ML package in ml/FormFitModel.mlpackage
  • Saved training curve in ml/training_history.png

Running The App

Open the Xcode project at:

watchos/FormFitWatch/FormFitWatch.xcodeproj

Then:

  1. Select the FormFitPhone scheme if you want the paired iPhone app flow.
  2. Select the watch app scheme if you want to work directly on watch collection.
  3. Run on a real iPhone + Apple Watch pair for sensor collection and transfer testing.

More detailed notes live in watchos/README.md.

Running The ML Pipeline

From the ml/ directory:

python3 -m pip install -r requirements.txt
python3 CNN.py

That script:

  • loads split rep CSV files
  • trains the 1D CNN
  • saves the best weights
  • writes a training history plot
  • evaluates validation accuracy
  • exports the model as a Core ML package

More details live in ml/README.md.

Current Status

This repo reflects a strong prototype milestone:

  • collection is working
  • watch-to-phone transfer is working
  • app-side storage is working
  • the UI is far beyond the original blank collector prototype
  • the ML training/export pipeline is present

The next major layer is tightening the connection between imported sessions and live in-app scoring / coaching.

Acknowledgment

FormFit was built through months of iteration across sensing, UI, data engineering, and model experimentation. This repository is meant to present that work as a cohesive system instead of only the earliest collector prototype.