This was an open competition to the members of the cohort to try and test their computer vision skills by implementing as a good a butterfly classifier as they could train within the time limits of the current term.

My approach to this task involved using a convolutional neural networks for fine-tuning a pre-trained model, which is a process called transfer learning. This allowed for my model to get a very high accuracy from the get-go while also having some room for further fine-tuning based on the CNN applied on top of it.

Using this method, I was able to make it into the top-10 classifiers in the cohort.

Transfer learning can be a very powerful and highly accessible tool for achieving good classification accuracy in a short amount of time.

Hyper-parameter tuning and and data-preprocessing can help to majorly boost generalisability of your model.