This project contains the code used to submit for the Eyes on the Ground competition on Zindi. Unfortunately, I was disqualified due to a change in rules that I wasn't aware of and didn't take into consideration, despite being ranked 8th on the public leaderboard.
Nevertheless, the competition was very enjoyable.
In this competition, the objective was to predict the extent of damage from crop images. The damage can be caused by drought, flooding, and the stage of growth can vary between the images.
This repository contains the model/dataset classes, training, inference, and submission generation codes. The experiments and training itself were conducted in a 2*T4 Kaggle notebook, with experiment tracking and model versioning in W&B.