Computer Vision Workshop

Train a neural network to classify images – with Raphael and Paul from cloudflight.

Idea

We want to train a neural network to find out if a smiley is happy or sad with hand-drawn similies.

Setup

• Setup a Python 3.8 environment - e.g. in PyCharm.
• Run pip install -r requirements.txt in the console to install tensorflow etc.
• Run jupyter lab to start jupyter lab and click the link in the console output to open it in a browser.
• Extract the data.zip containing all training and validation data.
• Run the notebook in the browser and change it as you like.

Result

We reached an accuracy of 92.5% on out validation set!

The validation set also contained some similies drawn by Raphael and Paul in Paint, and also some emojies from the internet - just as a showcase that the real word smilies can be different than what you would expect :)

Thank you!

Thank you so much for all the smilies! We did the workshop twice and have 291 smilies in our train set now! You are awesome!

Contact

If you have any questions, don't hesitate to write me on

• linkedin: https://www.linkedin.com/in/paul-puntschart-279506a2/
• email: paul.puntschart@cloudflight.io