YoVDO

FIFA 22's Hypermotion - Full-Match Mocap Driving Machine Learning Technology

Offered By: GDC via YouTube

Tags

GDC (Game Developers Conference) Courses Game Development Courses Machine Learning Courses Animation Courses Motion Capture Courses

Course Description

Overview

Explore the groundbreaking process behind FIFA 22's Hypermotion technology in this 2022 GDC Animation Summit talk. Delve into Electronic Arts' journey to capture a full 90-minute, 11-player vs. 11-player football match using motion capture technology. Learn about the challenges faced, innovative solutions developed, and the implementation of machine learning techniques. Discover the intricacies of optical motion capture, XSens prototype testing, and solving drift issues. Gain insights into the shoot requirements, camera setup, and workflow for finding animations. Understand the complexities of ball tracking, camera calibration, and physics tuning. Examine the machine learning runtime, input sources, and training data pipeline. Analyze stress test results, identifying successes and areas for improvement. Uncover new challenges and future directions for this revolutionary approach to sports game animation.

Syllabus

Intro
The Inspiration
The Problem Optical Motion Capture
XSens - Prototype
Problems to Solve
Solving Drift
Shoot Requirements
Camera Setup
During the Shoot
Workflow – Finding Animation
Ball Tracking
Tracking Pipeline
Camera Calibration
Tuning Physics
Generic ML Runtime
Inputs/Outputs
Source of Inputs
Data Source - XSens
Training Data Pipeline
Training Deep Neural Network
Identifying Problems - Stress Test
Stress Test - Reviewing Results
What Worked
What Needs Work New Challenges
In Conclusion
Other Sessions
We're Hiring


Taught by

GDC

Related Courses

Interactive 3D Graphics
Autodesk via Udacity
Creative Coding
Monash University via FutureLearn
Make Your Own 2048
Udacity
An Introduction to Interactive Programming in Python (Part 2)
Rice University via Coursera
HTML5 Coding Essentials and Best Practices
World Wide Web Consortium (W3C) via edX