Rethinking Fixed Temporal Grids: Frequency-Disentangled Motion Generation

Benchmark Model Rank Results
motion-synthesis-on-humanml3dFreqMo–FID: 0.040
motion-synthesis-on-humanml3dFreqMo-Diff–FID: 0.043
motion-synthesis-on-kit-motion-languageFreqMo–FID: 0.181
motion-synthesis-on-kit-motion-languageFreqMo-Diff–FID: 0.174