The article presents DanceMeld, an innovative dance generation pipeline for 3D digital human applications. It uses a hierarchical VQ-VAE to disentangle dance poses and dance movements. This allows the system to generate dance sequences that match the rhythm, melody, and style of the music. The system has been successfully tested on the AIST++ dataset, demonstrating its superiority over other methods. DanceMeld has potential applications in the gaming industry, film production, and AR/VR experiences, offering a cost-effective means for individuals to produce and edit dance sequences for 3D characters.
Publication date: 30 Nov 2023
Project Page: https://humanaigc.github.io/dance-meld/
Paper: https://arxiv.org/pdf/2401.10242
