The best large-scale deep learning models require massive amounts of training data. For some tasks, collecting such data may be unfeasible, due to logistical or legal reasons. Few-Shot Learning has emerged as a field of study to maximise performance based on very few samples. In Generalised Few-Shot Learning, a model has to learn new few-shot classes while recalling earlier large-scale training classes. In this work, we review the tools and techniques used in Few-Shot Learning before exploring the parallels between Generalised Few-Shot Object Detection (G-FSOD) and Continual Learning (CL) methods. We focus on the manipulation of gradient descent since it has been recently proposed for G-FSOD. We show that gradient methods appear to be no better than existing techniques, and point out that potentially beneficial insights on sampling from the Continual Learning world have yet to be employed. We hope this work will
provide a blueprint for further study of both G-FSOD and CL as interconnected fields.
| Date of Award | 11 Jun 2025 |
|---|
| Original language | English |
|---|
| Awarding Institution | - University Of Strathclyde
|
|---|
| Supervisor | Marc Roper (Supervisor) & Andrew Abel (Supervisor) |
|---|