This is a project assignment for course WID3008 Image Processing. The objective of the project is as follow:
- To classify whether a given chest x-ray image is infected by COVID-19 or is a normal chest x-ray image
- To evaluate the accuracy of COVID-19 classification for chest x-ray images
More detailed information at the Jupyter Notebooks
pip install -r requirements.txt
or conda env create -f conda_env.yml
Datasets are obtained from Kaggle
Below are a few training examples:
The preprocessing steps are defined as follow:
Below is a few images comparison showing the output of CLAHE:
We use the InceptionV3 base model and add our own classifier dense layer to it to predict our binary image class. Training steps are as follow:
- Freeze base InceptionV3 model and train network head with Adam optimizer
- Unfreeze a few inception block and train using SGD with low learning rate