Publication

Abstract : Historically Deep learning algorithms like Face Recognition do not work well on One-shot learning tasks. Hence, we need algorithms that can learn information about objects from one, or a few, training examples or images. In addition to that, if any new class is introduced to the model, there will be requirement to train it from beginning. We solve these problems by using the weights of a pre-trained model. The goal of this project is to use the concept of one-shot learning to create a component for an offline proctoring system. This system is foolproof against seat-swapping and it is also resilient to different lighting conditions and poses. In this project, we have used a pre-trained siamese network model and have established the feasibility of the product on a number of metrics including the accuracy of the model under different conditions like change in orientation and lighting.