Last Sunday, I took a beginner’s session on getting started with artificial intelligence as part of AI-Creatives meetup Code & Coffee. To get the fancy of attendees, I picked up computer vision; more specifically, image recognition using deep learning.
As time was deliberately limited (90 mins), I focused on inference more than training. Deep learning, as you know, requires time (weeks to months) and resources (specialized GPUs). I based my demo on the Tensorflow’s image retraining tutorial.
Celebrity Recognizer Website
The core idea was to quickly retrain a trained ImageNet model, and then create an API around it for inferences. This API would then be tested by using it on a webpage. I think the session went well. You can see the deck I used here. Source code for classifier, API and webpage is on GitHub.
By the end of it, we weren’t able to create the website part but we did successfully complete the API part. Creating the API using Flask was fun; it’s my first Flask application.
I just finished the webpage. You can see how it looks in the screenshot above. I used Bulma, something I had been meaning to use for more than a week now.
P.S. Shivam, it’s finally here: a Bulma-based webpage. It’s very basic, I know, but it’s all pure Bulma.