In this project, I've used VGG16 architecture and used transfer learning to help ahieve a better accuracy.
There were 5216 training images and 16 validation images in the dataset from Kaggle. Using this model I've achieved 88.6% accuracy without
fine-tuning of the model.
And after fine-tuning ,i.e., unfreezing the frozen layers of VGG16 architecture and using very low learning rate to train the model again which helped me achieve 90.3%.
1.) Numpy.
2.) Keras.
3.) Pandas.
4.) Google Colab.
- Link to data-set: Kaggle Data-set .
- Link to google colab notebook : Transfer Learning Pneumonia .