Project for computer science course implementing triplet loss learning. Through this project I demonstrate my ability to use pytorch for image classification using AlexNet, transfer learning, from-scratch networks, and from journal descriptions. I also use industry-standard train, test, and validation data splits for accuracy reporting.
Through a series of problems, I compare the performance of several convolutional neural networks (CNNs). Throughout I use the NWPU-RESISC45 dataset (Northwest Polytechnic Institute: REmote Sensing Image Scene Classification).
In problem 1 I use a pre-trained AlexNet, which I then modify through transfer learning to increase the test accuracy by 5%.
In problem 2, I generate my own CNN in pytorch.
In problem 3, I take a journal paper which describes triplet margin loss, create the architecture in pytorch, and implement on our dataset.
