Self-supervised deep learning for missing image predictions

• Main Author: Yeoh, Shun Bin
• Other Authors: Yeo Chai Kiat
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/166706
• Institution: Nanyang Technological University

Deep learning has revolutionised the field of computer vision, enabling impressive advances in image classification, object detection, and segmentation. However, the success of supervised deep learning in computer vision is largely dependent on the availability of large annotated datasets, which can be time- consuming and expensive to acquire. Hence, self-supervised deep learning can be automated even with enormous volumes of unlabelled data, which reduces data labelling costs. This project explores self-supervised learning techniques for computer vision-based tasks, with a specific focus on patch context prediction for missing image predictions. We explore the potential of self-supervised learning to develop a model that can predict missing parts of an image based on the surrounding context and evaluate the performance of state-of-the-art (SOTA) techniques in this area. The goal of this report is to demonstrate the potential of self-supervised learning for practical image restoration scenarios and to provide insights into the impact of hyperparameters on model performance.