Stochastic human motion prediction using a quantized conditional diffusion model

Biaozhang Huang, Xinde Li, Chuanfei Hu, Heqing Li

Abstract: Highlights•Combines motion quantization with conditional diffusion model for motion prediction.•Utilizes GCNs and temporal convolutions for efficient motion feature extraction.•Integrates action category info for controllable, diverse motion generation.•Implements diversity enhancement to reduce similarity in prediction samples.