Self-Supervised Learning Research (2020)

In this project, I conducted research to identify the most effective architecture for learning robust feature representations from unlabeled images. I experimented with cutting-edge contrastive learning methods, including SimCLR (developed by Google) and Swav (developed by Facebook), leveraging various backbone networks such as ResNet, ResNext, and EfficientNet. These architectures were subsequently utilized for downstream tasks like image retrieval and object detection, where they demonstrated strong performance.