Training an Image Segmentation Model

From the description of the task:

There are 30 satellite pictures of houses and 25 corresponding labels that indicate the roofs. Take those 25 data points and train a neural network on them - you are completely free about the architecture and are of course allowed to use any predefined version of networks, however, you should be able to explain what you are doing - in terms of code as well as in terms of why certain steps are good choices. The preferred language is Python, but you can also use other languages. Please evaluate your network on the 5 remaining test images by making predictions of the roofs - send us the predictions and ideally some comments on what you have been doing. Everything else we will discuss from there.

Here you can view my model training scripts for this task. Additionally, you can view them on my GitHub.

  1. Initial baseline - I fix initial choices for the segmentation model, write a training loop, and define evaluation across various metrics. I also show how to make predictions on the unlabeled images.
  2. Iteration and experimentation - I perform experiments to iteratively improve upon the modeling choices. In order, I experiment with the model encoder and pretrained weights, the model architecture, the optimizers, and the loss function. I also defined a rudimentary dashboard to interactively plot these results and summarize my experimental findings.
  3. Final training and augmentation test - Having fixed informed modelling choices, this notebook contains the final training of the model. I also perform one final experiment of a control model, and a model trained on augmented data from the dataset, attempting to mitigate our original datasets small size. The control performed better, and so I made the final predictions using this control model.

For background information on this task, I learned a lot from Damek Davis’s notes A Playbook for Tuning Deep Learning Models and the write-up of the first place winners of the Kaggle competition Carvana Image Masking Challenge.