SMGDiff: Soccer Motion Generation using diffusion probabilistic models

1ShanghaiTech University, 2ByteDance
*Indicates Equal Contribution

Our method, SMGDiff, enables users to control soccer motions based on character displacement and soccer skill, simulating an interactive gameplay experience. It can generate a diverse range of high-quality soccer motions while ensuring real-time performance.

Abstract

Soccer is a globally renowned sport with significant applications in video games and VR/AR. However, generating realistic soccer motions remains challenging due to the intricate interactions between the human player and the ball. In this paper, we introduce SMGDiff, a novel two-stage framework for generating real-time and user-controllable soccer motions. Our key idea is to integrate real-time character control with a powerful diffusion-based generative model, ensuring high-quality and diverse output motion. In the first stage, we instantly transform coarse user controls into diverse global trajectories of the character. In the second stage, we employ a transformer-based autoregressive diffusion model to generate soccer motions based on trajectory conditioning. We further incorporate a contact guidance module during inference to optimize the contact details for realistic ball-foot interactions. Moreover, we contribute a large-scale soccer motion dataset consisting of over 1.08 million frames of diverse soccer motions. Extensive experiments demonstrate that our SMGDiff significantly outperforms existing methods in terms of motion quality and condition alignment.

Pipeline

MY ALT TEXT

Our framework consists of two stages. In the first stage, the trajectory encoder will first transform coarse user controls into detailed human trajectory, and the trajectory blending ensuring the smoothness of the trajectory. In the second stage, the soccer motion diffusion model will generate soccer motion based on the trajectory conditioning, a contact guidance module will optimize the contact details for realistic ball-foot interactions.

Soccer-X Dataset

MY ALT TEXT

To train and evaluate model performance, we construct Soccer-X, a high-quality human-soccer interaction dataset covering diverse soccer motions. It covers six common soccer motion categories, comprising over 1.08 million frames and more than 10 hours of diverse and high- quality motion data from 30 soccer players.

Dribble

Stand

Celebrate

Trick

Shoot

Off-the-ball Move

Results

Shoot

LMP

MANN-DP

CM

Ours

Trick: Marseille Turn

LMP

MANN-DP

CM

Ours

Trick: Step Over

LMP

MANN-DP

CM

Ours

Dribble

LMP

MANN-DP

CM

Ours

Citation


        @misc{yang2024smgdiffsoccermotiongeneration,
          title={SMGDiff: Soccer Motion Generation using diffusion probabilistic models}, 
          author={Hongdi Yang and Chengyang Li and Zhenxuan Wu and Gaozheng Li and Jingya Wang and Jingyi Yu and Zhuo Su and Lan Xu},
          year={2024},
          eprint={2411.16216},
          archivePrefix={arXiv},
          primaryClass={cs.CV},
          url={https://arxiv.org/abs/2411.16216}, 
      }