Transfer-Learning-for-Computer-Vision

This project aims to compare some of the existing works in transfer learning, touching various applications. We evaluate performance of using pretrained networks, domain adaptation, model-agnostic meta-learning, and few shot learning. We are evaluating the methods on the following dataset pairs - MNIST and Fashion-MNIST, CIFAR-10 and CIFAR-100. Domain adaptation would be evaluated on Office-31 dataset. For MAML, sinusoid regression was chosen and synthetic data points were generated.