Reproduce “FFMT: Frequency-Factorized Multi-Rate Tokenization for Music-to-Dance Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Qirong Guo, Ruijie Li, Zeyan Li, Sicheng Shu, Guanzhi Ding, Weihang Ding
FFMT separates root-relative joint motion using a fixed Gaussian filter into a low-frequency scaffold and high-frequency residual, then encodes them at different token rates and codebook sizes. It generates the scaffold from music before predicting detail conditioned on music and scaffold. On AIST++ and FineDance it reports lowest FID among compared methods, but not closest ground-truth match across all other metrics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.