Classifying infected vs. uninfected malaria cells.
CNN from scratch is just a 3 convolutional layer and 2 fully connected layer cnn. It is small but when I tried a larger one it didn't work as well.
I used a vgg16 model for transfer learning. I changed the output nodes to 2 and got 78% accuracy for testing and training.
I started with a learning rate of 0.01 for 20 epochs. Then when my model had trained I went down to about 0.003 and so on
for about 4 more cycles of that and got around 70% acurracy on the training data.
I used this kaggle dataset. Requirements are Pytorch and Numpy
I used google Colab GPU because I don't have a good computer. It trained 20 epochs in around 10 minutes.