Multimodal DuetDance (MDD)
收藏资源简介:
MDD是一个多模态数据集,旨在为文本控制和音乐条件下的3D双人舞蹈动作生成提供支持。该数据集包含由专业舞者表演的620分钟高质量动作捕捉数据,与音乐同步,并附有超过10K个细粒度的自然语言描述。这些注释捕捉了丰富的运动词汇,详细描述了舞伴之间的空间关系、身体运动和节奏,使MDD成为第一个无缝集成人体运动、音乐和文本以生成双人舞蹈的数据集。MDD支持两项新任务:Text-to-Duet和Text-to-Dance Accompaniment,分别用于生成协调的舞伴动作和根据领舞者的动作生成跟随者的动作。
MDD is a multimodal dataset designed to support 3D duet dance motion generation under text control and music conditioning. This dataset contains 620 minutes of high-quality motion capture data performed by professional dancers, synchronized with music, and accompanied by over 10,000 fine-grained natural language descriptions. These annotations capture a rich motion vocabulary, detailing the spatial relationships between dance partners, bodily movements and rhythms, making MDD the first dataset that seamlessly integrates human motion, music and text for duet dance generation. MDD supports two novel tasks: Text-to-Duet and Text-to-Dance Accompaniment, which are respectively used to generate coordinated duet partner motions and generate the follower's motions based on the lead dancer's movements.



