Self-supervised learning for animation

In response to the dynamic challenges inherent in the gaming industry’s character animation pipelines, characterized by the labor-intensive and error-prone nature of manually labeling frame-by-frame data, this project endeavors to integrate Self-Supervised Learning (SSL) techniques into the animation workflow. SSL offers a compelling solution by enabling the acquisition of efficient and generalizable representations from large sets of unlabeled data, thus mitigating the substantial demand for accurately labeled data. The principal objective is to harness SSL to obtain a robust representation for full-body human motion, subsequently deploying this learned representation across animation-related tasks relevant to the interests of La Forge and Ubisoft. This innovative approach, marked by increased efficiency, accuracy, and versatility, promises to revolutionize character animation pipelines within the gaming industry, ushering in a paradigm shift with a reduced reliance on annotated data.

Faculty Supervisor:

Konstantinos Derpanis

Student:

Partner:

Ubisoft Divertissement;Ubisoft Toronto

Discipline:

Computer science

Sector:

Information and cultural industries

University:

York University

Program: