Opticus

• 

  Landing page

• 

  Input

• 

  Output

• 

  Facial Landmark Detection

• 

  Semantic Segmentation of Piece of Paper

Inspiration

Our friend had complain that they had trouble buying glasses online that fit properly and was confused by the measurements that needed to be taken, and had trouble taking the measurements herself.

What it does

Calculates your ideal glass size including frame width, lens width and arm length from your picture where you're holding a piece of paper.

How we built it

The front end was built using React + Ant.Design.

The back end is a Flask app that is running 2 machine learning computer vision models. The first model does landmark recognition on your face to determine where your eyes and sides of your head are and then calculates pixel distance of your temple width and nose bridge width. To get an actual measurement, we then do semantic segmentation to find the piece of paper in the image and use the known standard lengths of the sides of the paper to make sense of the pixel distance found in the previous model.

Challenges we ran into

Deploying machine learning models was a first for all of the members of the team and was challenging to complete in the allotted time.

Accomplishments that we're proud of

Hopefully finishing on time!

What we learned

We learned a ton about how ML models get deployed and how to do land mark recognition on faces as well as semantic segmentation to detect the piece of paper.

What's next for Opticus

Refining our models to give more accurate output! This was design to be a proof of concept application and with some more work we believe we could get even better accuracy!

Built With

• flask
• python
• pytorch
• react