This is a Convolutional Neural Network (CNN) based ensemble model - CovidNet to classify the X-Ray radiographs into one of the 3 categories : Covid-19, Pneumonia (Viral) and Normal. I have build the ensemble model using 2 parallel pretrained EfficientNet-B0 as backbone networks and averaged their results to classify the High Resolution RadioGraphs accurately reducing the variations in predictions. The dataset is taken from the COVID-19 Open Research Dataset and consists of about 1300 Covid, 1341 Viral and 1355 Normal High Res Radiographs. Link : https://www.kaggle.com/tawsifurrahman/covid19-radiography-database
The model outperforms the state-of-the-art models trained before achieving about ~95% accuracy and loss of about 4% on the test-set consisting of 120 random radiographs each from 3 classes.
Sample Predictions are given below :
