Using mobiles to identify errors in people performing exercises with deep learning - BC-462

Desired discipline(s): Engineering - computer / electrical, Engineering, Computer science, Mathematical Sciences
Company: Flex AI
Project Length: 6 months to 1 year
Preferred start date: As soon as possible.
Language requirement: Flexible
Location(s): Vancouver, BC, Canada
No. of positions: 1
Preferred institutions: Simon Fraser University, University of British Columbia, University of Waterloo

About the company: 

Creating a physically healthier world!

We are Vancouver’s fastest growing startup, having secured $2M in funding since being founded last year. We have an extraordinarily talented team coming from all sorts of formidable backgrounds.

Our mission is to put a personal trainer in everyone's pocket and to create a physically healthier world. We doing this by using mobiles to identify errors in the form of people performing exercises with deep learning.

We're building a simple, elegant, fun app that incorporates our cutting-edge patented A.I. to help everyone improve their fitness.


Please describe the project.: 

While performing an exercise in the gym or at home, it is crucial to have the right body posture for maximum benefit and injury avoidance. This is especially important when performing exercises with heavy weights. In order to monitor the body posture, one constantly needs an expert closely monitoring the exercise which may not be practical for most people.

To solve this, we are building a system to work on mobiles thatcan monitor posture of the person performing exercises and provide immediate feedback if there is an error in the body posture or in the way an exercise is performed.

For this project we are building custom accurate datasets labeled by expert annotators. We use state-of-the-art deep learning model
architectures for image and video classification, human pose estimation, object detection to analyze the videos. Further, to run these deep
learning models on mobiles we are exploring ways for model compression to reduce computation cost and improve speed while
maintaining model performances.

Academic researchers with graduate students working in AI/deep learning are encouaged to contact us to discuss this exciting research project and proposed methodologies. 

Required expertise/skills: 

- Experience with Python, deep learning frameworks like keras,tensorflow or pytorch, opencv.

- Experience with deep learning model compression and deployment on mobile and embedded devices.

- Experience with computer vision algorithms and deep learning models for image classification, video classification, human pose-estimation, object detection, segmentation.