MoDiPO: text-to-motion alignment via AI-feedback-driven Direct Preference Optimization

arXiv:2405.03803v1 Announce Type: new
Abstract: Diffusion Models have revolutionized the field of human motion generation by offering exceptional generation quality and fine-grained controllability through natural language conditioning. Their inherent stochasticity, that is the ability to generate various outputs from a single input, is key to their success. However, this diversity should not be unrestricted, as it may lead to unlikely generations. Instead, it should be confined within the boundaries of text-aligned and realistic generations. To address this issue, we propose …

abstract alignment arxiv cs.cv diffusion diffusion models direct preference optimization diversity feedback fine-grained generate however human key language natural natural language optimization quality success text through type via