This repository contains our code for the AMIA Kaggle Challenge as part of the "Computer vision for biomedical images" seminar from Sören Lukassen @FU Berlin.
We implemented some preprocessing steps to analyse the images and their labels, before doing a cluster analysis of the bounding boxes to obtain adequate anchor boxes for training different object detection models.
The goal of this project was predicting up to 14 different findings in chest X-ray images.
We trained FasterRCNN and RetinaNet with different hyperparameters to find the best model for the given dataset.
A full report on how we tackled this project can be found here (Max' report).
Alternatively, an HTML version of the project report can be found here (Flo's report).