This is a workout assistant that can count the reps of workouts in realtime using pose-estimation.
- A convolutional neural network is trained on pose data of a selected exercise (shoulder dumbbell press) in it's various states.
- Pose data is generated from live webcam feed
- Classification is performed on it to classify it into one of the several states of an exercise.
- Based on this classification, reps are counted.
The user can view the results in a python app.
This project is currently a WIP. I am working along the following timeline:
- Get train/test images
- Write tools for automated pose-detection in input images
- Train a CNN model with acceptable loss on pose data
- Do local evaluations, re-trainings
- Make GUI
- attempt deployment