A new framework for the evaluation of locomotive motion datasets through motion matching techniques

What is it about?

Imagine you need to make a virtual character move. You would try finding a way to move its arms, legs, torso and find out that this can be accomplished through animation. You then start searching for animations that fit your purpose, such as one that makes the character's hand wave at something and so on. However, this is not enough, you now want to make it run. You find many animation datasets on the internet that can suit your needs, but how can you tell them apart? How can you say that one of the datasets can do more than the other? In this work, I introduce a way to solve this problem.

Featured Image

Photo by Ryan Garry on Unsplash

Photo by Ryan Garry on Unsplash

Why is it important?

This work is the first attempt at standardizing the evaluation of motion datasets. Most animators and the animation research community will say that the most important thing when it comes to animating characters is the animations you have available, not the animation selection algorithm, but still there is no standard way of looking at motion datasets, leaving this reasoning to the subjectivity of the evaluator.