model both style and variation in human motion (1st sentence, 2nd paragraph)
2. Problem: Individually capturing the motions for all characters in a
high density crowd animation is impractical, which leads to the use of distracting motion clones.
3. Key Idea: Model Style and variation separately using latent variation parameters. This guarantees that even for the same style (e.g. stride length of 1m) there is variation that still looks natural
4.a What the paper does: Uses example motions (side stepping, walking, running) and parametrizes them on some simple styles (stride length, speed). It then generates variations of these styles that fall within the constraints, but look different and natural.
4.b. What it could be used for: We could couple this with motion graph structures and larger types of animations to create realistic variations in environments with large number of virtual characters. This might also apply to video games when a player is always following the same character.
5. Resources:Video