The purpose of the thesis is to investigate the potential of using diffusion models as a tool for Manga colorization. In addition to selecting the optimal architecture and parameters, the work analyzes the dataset with Manga, its generalizing ability, and in general various methods of comparing and benchmarking models for colorization. GitHub Repo

The advancement of machine learning models for sequence data has seen significant progress with the development of the Structured State Space Sequence (S4) architecture. Initially designed for natural language processing (NLP) tasks, S4 has demonstrated superior performance over traditional transformers, particularly in handling long-context dependencies. While transformers have successfully migrated from NLP to computer vision, becoming the state-of-the-art in various tasks, they still face challenges with long-context information. This thesis explores the potential of S4 in the realm of computer vision by adapting it to high-dimensional data. The primary focus is on evaluating S4ND in computer vision tasks involving high-resolution images and long-context information. GitHub Repo