We read an article by Nvidia that successfully used neural nets to slow down animations by adding in extra frames. We thought the concept of having AI generate their own frames was quite cool. Was it possible to take two key frames, and have a neural network generate frames in between?
What it does
The user will draw a series of frames. They would ideally be frames of an animation that are a few seconds apart. Our neural network will take these frames as input and complete the animation for you
How we built it
UI/Frontend: Built using React Algorithm: Python, Tensorflow
Challenges we ran into
Training a neural net is very time consuming on CPUs. We had to purchase GPU servers. The neural net was also very difficult to create, as frame generation is a topic currently still in the research phase, which meant we were working at the boundary of machine learning intelligence.
Original research paper this project was based off on by Cornell University: https://arxiv.org/abs/1702.02463