The project tests two pretrained models, namely ResNet18 and VGG16 on classifying the CalTech Birds Dataset.
The project is an attempt to understand the importance of transfer learning. Transfer Learning is incredibly crucial when we are posed with the problem of a small dataset. Hence we use models pre-trained on the larger ImageNet dataset, modify only their input and output layers and freeze their hidden layers. This saves time during training and still gives decent performance.
nntools.py -- Contains abstract classes for implementing various networks, calculating performance,etc.
main.py -- Code for training and testing.
tutorial.pdf -- A JuPyter Notebook file for explanation purposes.