TM2T: Stochastic and Tokenized Modeling for the Reciprocal Generation of 3D Human Motions and Texts

Benchmark Model Rank Results
motion-captioning-on-humanml3dTM2T#3BLEU-4: 22.3BERTScore: 37.8
motion-captioning-on-kit-motion-languageTM2T#3BLEU-4: 18.4BERTScore: 23.0
motion-synthesis-on-kit-motion-languageTM2T#25FID: 3.599R Precision Top3: 0.587Diversity: 9.473…
motion-synthesis-on-kit-motion-languageLanguage2Pose#27FID: 6.545R Precision Top3: 0.483Diversity: 9.073
motion-synthesis-on-kit-motion-languageText2Gesture#28FID: 12.12R Precision Top3: 0.338Diversity: 9.334
motion-synthesis-on-humanml3dTM2T#38FID: 1.501R Precision Top3: 0.729Diversity: 8.589…
motion-synthesis-on-humanml3dText2Gesture#39FID: 5.012R Precision Top3: 0.345Diversity: 6.409
motion-synthesis-on-humanml3dLanguage2Pose#40FID: 11.02R Precision Top3: 0.486Diversity: 7.676