Week 2 review

We hope that it has given you some useful ideas as to the best ways to organise and maximise the use of your datasets.

Specifically we looked at the following:

• organising your datasets
  • sensible filenames – unique and descriptive names can help to organise your data
  • use of QR codes and barcodes as labels within images
  • scales within images – for relating pixels to actual size
  • division of datasets into training, valdation and test sets
• expanding your datasets
  • augmentation – making slightly altered copies of existing images
  • synthetic data – creating entirely new, artificial images to train models
  • pre-existing datasets – other studies may have used similar images with which you can partially train your model
• releasing data
  • types of license – e.g. Creative Commons
  • reasons for sharing data.

What’s next?

In the first two weeks of the course we have focused almost entirely on the data aspect of machine learning. In Week 1 we talked about acquiring and annotating data, while in Week 2 we looked at ways in which data can be organised and supplemented.

Next week, we will begin looking in more detail at what we can do with our data using machine learning and deep learning models.