Gaming, filming, and animation are typical applications for FlowMDM, where text is a very natural, powerful, and user-friendly way to guide motion generation and control motion semantics. “Let’s say you’re preparing a storyboard and imagining how things will look,” Cristina says. “You can use text-tomotion. First, a person will start walking, and then they will kick a ball with their right foot. You can see this motion. If you can visualize things, you can better appreciate if you like them or not.” FlowMDM has potential applications in many other areas, including human-computer interaction, physical training, and robotics. “We could be guiding a robot in surgery, for example,” German points out. “We know that it needs to extract an organ, and we need to generate the transitions between extracting the organ and putting it in someone else. Everything is guided by generating transitions. It’s a really central problem.” The team’s greatest challenge was 20 DAILY CVPR Wednesday Poster Presentation
RkJQdWJsaXNoZXIy NTc3NzU=