Course review

We’ve covered a lot during the course, in what is a relatively new and fast moving topic, but hope we have given you a flavour of what deep learning is, and what it can do.

In the worked practical examples we hope we have demonstrated that with relatively few lines of code, you can quickly build up very powerful and effective tools for a range tasks useful for plant phenotyping, plus many other fields of study.

In particular we have built an image classifier, a regression model for counting flowers in an image, and a segmentation model to outline flowers within images.

Lets just recap what the course has covered:

• deep learning vs machine learning
• Python deep learning libraries – in particular PyTorch
• convolutional neural network (CNN) components
  • convolutional layers
  • max pooling layers
  • activation functions
  • tensors
• building a CNN in PyTorch
• training CNNs
• improving training by adjusting hyperparameters
• datasets and dataloaders – managing your data efficiently
• encoder architctures for classification and regression
• encoder-decoder architectures for image segmentation
• more advanced topics
  • heatmap regression
  • bounding box methods
  • multi-task learning

Quite the list! Though we’ve only really scratched the surface of these topics, you should at least have a clearer idea of what they are and the terminology used when discussing them.