Abstract
The current animation process is repetitive and labor intensive for the animators since the standard for animated videos is 24 frames per second. That means the animators have to draw 24 images for 1 second of motion. Frame interpolation reduces this labor by taking two images as an input and creating possible images that fit in between the two input images. It is still a topic of active research and improvement, but tools for animation are currently restrictive or lacking. The two key issues with the existing frame interpolation methods are unusual smoothness and limited scope of simple methods, and blending or distortion of the in-between frames in machine learning based methods. Our goal is to create an acceptable quality shortcut for independent creators, using a relatively simple model combining flow estimation and stroke identification techniques. Using the method proposed, the model in this paper can generate in-between sketch frames with a higher pixel to noise ratio compared to two of the baseline models presented in this work AnimeInbet and VFIT.