less than 1 minute read

Foo

At the end of the computer vision course, this competition was recommended. There was a notebook aready completed which could be forked and then edited to improve the score (initially 0.04265). My edited version, which was submitted to the competition, can be found here.

Changes I made to improve the score:

  • Added stopper function
  • Changed pretrained model to Xception
  • Adjusted the training schedule
  • Added augmentation to the dataset (before it is fed to the model), however this slows it down too much so is not activated right now

The final score is 0.94012 which is 26th on the leaderboard (as of 05/10/2022).

Foo

As of 18/10/2022, this submission has been awarded (my first) bronze medal.