Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore a comprehensive technical analysis of Lumiere, a space-time diffusion model for video generation, in this 43-minute deep-dive video. Learn about the challenges faced in current video generation techniques and how Lumiere addresses these issues through its innovative approach. Understand the comparison between Pepperoni Hug Spot and Lumiere, delve into diffusion probabilistic models, and examine the full inference pipeline. Discover how Lumiere preserves natural motion through its STUNet architecture and MultiDiffusion implementation. The presentation covers various downstream applications, evaluation metrics, and concludes with key takeaways for implementing this cutting-edge video synthesis technology. Perfect for AI researchers, developers, and anyone interested in the latest advancements in video generation using diffusion models.
Syllabus
Intro to Lumiere
Problems with Current Video Generation
Pepperoni Hug Spot vs. Lumiere
Problems with Approaches Before Lumiere
Lumiere’s Solution
Diffusion Probabilistic Models
Diving Into Lumiere
Full Inference Pipeline
Preserving Natural Motion
MuliDiffusion
STUNet Architecture
Downstream Applications
Evaluation
Takeaways
Taught by
Oxen