I came into the hackathon looking up various papers on Google's research pages.

Original Idea

I wanted to create a classifier which can tell whether a video is animated or live-action and then create an appropriate title for the video clip.

Starting out simple

To start out with a simpler form, I focused on just creating a classifier which takes an image and classifies it as either "animated" (0) or "live-action" (1).

Getting the datasets was hard so instead... I made a script which executes selected google searches and crawls to resulting websites for the images I needed to train and test on.

One of the mentors suggested I learn how to use Keras so I worked on using Keras to create the network and get a simple classifier working.

Built With

  • 5-hours-of-sleep
  • keras
  • pillow
  • python
  • tensorflow
Share this project: