SIIM-FISABIO-RSNA
Computer Vision
Object Detection
CNNs
fastai
icevision
YOLOv5
Competition
Identify and localize COVID-19 abnormalities on chest radiographs.
In this project I prepared a medical image dataset and the related dataframe to be used in a Neural Net that predicts an arbitrary number of labeled bounding boxes. These boxes are designed to assist doctors in identifying COVID-19 infections by suggesting an area of interest and a classification of them. The backbone of this project is a RetinaNet, that makes use of Anchor boxes and Focal Loss and can be used with different encoders like (X)Resnets or EfficientNets.