Dyadic Motion Prediction

Collect recent works on dyadic human motion prediction (multi-person human motion predction) and datasets which contain multiple persons

If you find some overlooked papers, please open issues or pull requests.

Paper

• Isinsu Katircioglu et al., "Dyadic Human Motion Prediction", arXiv 2021.12 [paper]

• Wen Guo et al. "Multi-Person Extreme Motion Prediction", arXiv 2021.05 [paper] [webpage] [code]

• Jiashun Wang et al., "Multi-Person 3D Motion Prediction with Multi-Range Transformers", NeurIPS 2021 [paper] [webpage]

• Vida Adeli et al., "Tripod: Human trajectory and pose dynamics forecasting in the wild", ICCV 2020 [paper]

• Vida Adeli et al., "Socially and contextually aware human motion and pose forecasting", RA-L 2020 [paper]

Datasets

• [ExPI]
  • in Multi-Person Extreme Motion Prediction
  • dataset: extreme dancing
  • 115 sequences, 16 actions, ~ 30k frames
  • collected by 68 camers, 3D poses infered from markers on the dancers captured by 20 mocap cameras
• [LindyHop600k] (not available yet)
  • in Dyadic Human Motion Prediction
  • dataset: dancing
  • 9 sequences, 4 actions, ~ 40k frames
  • collected by 8 cameras, 3D poses infered from OpenPose
• [CHI3D] (not available since CVPR 2020)
  • in Three-dimensional Reconstruction of Human Interactions
  • dataset: lab-based accurate 3d motion capture
• [3DPW]
  • in Recovering Accurate 3D Human Pose in The Wild Using IMUs and a Moving Camera
  • dataset: people in the wild
  • 60 sequences, 5 scenarios, ~ 50k frames
  • collected by a moving phone camera, 3D poses infered from their paper
• [MoPoTs]
  • in Single-Shot Multi-Person 3D Pose Estimation From Monocular RGB
  • dataset: in door and out door scenarios
  • 20 sequences, ~ 8k frames
  • data and 3D poses are obtained from a multi-view marker-less motion capture system Captury