This project aims to evaluate online continual learning methods on image datasets. Online continual learning studies how machine learning models can learn continuously from sequentially arriving data while reducing catastrophic forgetting of previously learned classes or tasks. This is important for adaptive AI systems that need to update their knowledge over time without retraining from scratch. The project will focus on image-based online continual learning experiments using deep learning models and standard evaluation protocols. Several methods will be compared, including replay-based approaches, exemplar-free approaches, and feature-based online classifiers. The experiments will be conducted under class-incremental or task-incremental settings, where new classes or data distributions arrive sequentially. The evaluation will consider accuracy, forgetting, memory usage, computational cost, and robustness across different data streams. The computational work will include image preprocessing, feature extraction using convolutional neural networks or pretrained image models, model training, online test-then-learn evaluation, repeated runs with different random seeds, and hyperparameter testing. The results will support the development and evaluation of continual learning systems that can adapt to changing visual data in a more efficient and reliable way. The project will mainly use Python, PyTorch, torchvision, NumPy, and scikit-learn. GPU resources are required for training and evaluating deep learning models on image datasets, while storage resources are required for datasets, preprocessed data, extracted features, model checkpoints, logs, and result files. Main supervisor: Zhibin Zhang, Uppsala University.