Automated colorization of animated characters

Abstract

This project undertakes the task of automating colorization in animations by exploring Frame-by-Frame Prediction Models and T-pose Reference-based Prediction approaches. Emphasizing the imperatives of reducing manual labor burden on Digital Painters, the study advocates for the adoption of innovative frameworks. The analysis delves into Frame-by-Frame Prediction Models, analysing performance of Segment Matching and Optical Flow through RAFT, each presenting its own merits and drawbacks. Additionally, image segmentation models, including PSANet and PSPNet, are investigated for possible integration into Segment Matching Models to achieve T-pose Reference-based Prediction. Moving forward, further research and development are crucial to enhance image segmentation methods and seamlessly integrate them into colorization workflows, ushering in automation in animation production.