Organic and believable Character Animation is increasingly important not only for Game Development but also for AR/VR applications and movie production (think about special effects, animating fantastical creatures and generating large scenes with thousands of characters, which would hardly be feasible using only real actors); however, animating by hand or using motion capture for every possible movement could be an extremely long, demanding and expensive task hence the development of techniques that use Machine Learning to generate Character Animations is more relevant than ever before.
During this seminar, we will explore together the evolution of the field during the last 10 years understanding various possible approaches to the problem (from the more data-intensive to the more training-based), then the participants will select a specific paper to focus on, understand in detail and present to the class and, after the presentations, there will be a time for discussing what has been learned and for comparing findings and opinions both on the theory and ideas behind the publications and on the implementation details, which will be part of the practical part at the end of the course and, hopefully, will lead to a Research Project in the following semesters.
Outline of the seminar (which could be modified):
* Introduction and Overview of the topics as frontal lectures (2-3 weekly meetings):
* What is Procedural Animation?
* Overview of the field and its common challenges/approaches/techniques
* Overview of some fundamental papers + interesting talks from conferences where important techniques (like Motion Matching) were presented by their creator
* Individual paper study and presentation preparation (2-3 weeks – no meetings)
* Student presentations of selected papers (1-2 weekly meetings)
* Discussion/Round-table with students about papers, for questions and exchange of ideas (1-2 weekly meetings)
* Implementation tools (Unity/Unreal + Tensorflow/PyTorch, and alternatives)
* Eventual Final Project: Tutorial-like ML Implementation with chosen tools |
