Inspiration

Many many companies are keen on finding the advertising tool that helps to find the perfect matches among the clients and their products. Amazon recommends product the clients might be interested at, YouTube suggests numerous videos based on people's taste.

Our impression was to help people advertising their product by introducing a new tool.

What it does

Emaze is a tool that captures facial expressions while the clients are observing a product or an application. It helps the investors to find which of their products delight others and which do not. Besides, it makes finding similar people more easy, because it provides a new feature space where people can be compared.

How we built it

Our dataset was downloaded from kaggle, and we applied learning algorithm on that, a neural network whose structure was inspired by Google's MobileNet. Using the neural network (after training) we kept generating probability distributions on the images our webcam captured.

Although the algorithm and our project were centralised around this matrix and this is our 'deliverable product', we thought it would be wise to visualise what was going on in the background, thus we connected our backend with a HTML webpage where we could present how the engine captures our reactions on a youtube video.

Challenges we ran into

  1. We expected better classification performance in the beginning
  2. Turned out it's not straightforward to how to connect python and javascript codes
  3. In the test phase we forgot that the clean training dataset was cropped and zoomed to actual human faces, while our camera captured the background, our neck and t-shirt. Thus, we needed to select human faces from the images we captured in live.

Accomplishments that we're proud of

  1. As it looks on cross validation sets, our final facial recognition model could achieve a top 20 rank on kaggle.
  2. We achieved this performance with laptops.
  3. We managed to both select faces from images (crop & select) and recognise emotions.

What we learned

  1. We should have agreed on the right idea earlier.
  2. We should have organised sleeping better

What's next for Emaze

We haven't thought anything special just yet.

Share this project:
×

Updates