Search:
Match:
1 results

Analysis

This paper addresses the limitations of existing text-to-motion generation methods, particularly those based on pose codes, by introducing a hybrid representation that combines interpretable pose codes with residual codes. This approach aims to improve both the fidelity and controllability of generated motions, making it easier to edit and refine them based on text descriptions. The use of residual vector quantization and residual dropout are key innovations to achieve this.
Reference

PGR$^2$M improves Fréchet inception distance and reconstruction metrics for both generation and editing compared with CoMo and recent diffusion- and tokenization-based baselines, while user studies confirm that it enables intuitive, structure-preserving motion edits.