Act As You Wish: Fine-Grained Control of Motion Diffusion Model with Hierarchical Semantic Graphs Supplementary Material

This appendix provides additional discussions (Sec. Although our method makes some progress, there are still many limitations worth further study. In this paper, we focus on improving the controllability of text-driven human motion generation. Node type Description Motion global motion description Action verb Specific attribute of action Edge type Description ARG0 agent ARG1 patient ARG2 instrument, benefactive ARG3 start point ARG4 end point ARGM-LOC location (where) ARGM-MNR manner (how) ARGM-TMP time (when) ARGM-DIR direction (where to/from) ARGM-ADV miscellaneous ARGM-MA motion-action dependencies OTHERS other argument types, e.g., action The overall sentence is treated as the global motion node in the hierarchical graph. Please refer to our code for more details.