I, like many others, have began exercising from home more and more due to quarantine, and I have noticed that sometimes when I am unsure how to correctly do an exercise or stretch, I end up in pain the next day after having done an exercise incorrectly. Without a person to spot your mistakes or interactive group fitness classes, where instructors offer insights to correct your form, exercising can lead to unwanted injuries and even chronic pain. I wanted to find a solution that would help correct my form and offer suggestions for improvement.
What it does
Downward Dog is an app that allows you to take a picture of your yoga poses, and the app will automatically identify the pose you are doing and provide a score (out of 100) for your pose. After receiving a score, users can proceed to an informational page for that pose and receive tips and feedback to improve their form and prevent injury based on their score.
How I built it
I built Downward Dog using CreateML and custom trained an image classifier model to identify and score different yoga poses. I incorporated the model in Swift and created an iOS app.
Challenges I ran into
The initial accuracy of my image classifier model was quite low, but after working with augmentation, I was able increase the testing accuracy.
Accomplishments that I'm proud of
Creating a custom image classifier model and incorporating that into an app.
What's next for Downward Dog
I hope to be able to incorporate AR to allow users to view demos of the poses, right in their own homes!