This project investigates the transferability of convolutional neural networks (CNNs) and transformer-based models. We will train both model types on the Describable Textures Dataset (DTD) and fine-tune them on the RSSCN7 dataset. Additionally, we will compare their performance with pretrained models from ImageNet and apply eXplainable AI (XAI) methods to interpret model decisions.
This part of the project was described in detail in the Presentations-Report/TransferLearning_ResearchProposal.pdf
To make a long story short, we decided to explore curriculum learning (self-paced method and with manual division of images into 3 parts) on Resnet18.
The whole project was described in detail in our report : Presentations-Report/Project-report.pdf