Speaker: Dominik Hanko (193-02 Computer Graphics)

The field of research in the use of neural networks to help artists or advance 2D animation is very underdeveloped. Most of the little research that exists does not ask questions that are relevant for animators but is done in a pure research mindset. We, however, tried to find a problem that would be relevant in the animation industry. Our idea is based on a cost and time-saving measure used in 2d animation, where features are often drawn in different levels of detail depending on factors like the current focus of a scene. This work aims to increase this level of detail for lower detailed features.
Previous achievements using state of the art image to image translation networks have shown some promising results in this field. These results, however, still suffer from problems regarding their color and scale when compared to the original image. To fix these problems we plan to acquire a sufficiently sized dataset by developing an automated solution that can extract the required data from animation frames. With this dataset, we then plan on further adapting state of the art image to image translation networks to fix the existing problems and generate higher detailed images that can be used by animators.




10 + 10
Supervisor: Josh Cardoso