EgoPoseVR: 18,235 Synthetic Motion Clips for Egocentric Full-Body Pose Estimation

EgoPoseVR is a large-scale synthetic dataset for egocentric full-body pose estimation in virtual reality. It contains 18,235 motion clips with paired RGB-D observations, pose annotations, HMD tracking signals, and SMPL body parameters across 7 virtual scenes. The dataset was created by AplusX and was last updated on April 12, 2026.

Use Cases

• Train egocentric full-body pose estimation models based on the paired RGB-D observations and pose annotations.
• Develop VR avatar systems based on the SMPL body parameters and temporally aligned motion clips.
• Research sensor fusion for motion tracking based on the HMD tracking signals and pose data.
• Benchmark pose estimation algorithms across different virtual environments based on the 7 provided scenes.

Strengths

• Contains 18,235 motion clips, providing a substantial volume of training data.
• Includes multiple synchronized data modalities: RGB-D, pose, HMD tracking, and SMPL parameters.
• Data is split into train (14,702), validation (1,827), and test (1,706) sets for machine learning workflows.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data is synthetic, which may limit its applicability to real-world scenarios without domain adaptation.

Provenance

Source

AplusX on Hugging Face.

Collection Method

Synthetically generated in virtual reality environments.

Time Range

null

Freshness

Last updated 2026-04-12 14:20:36; freshness should be verified.

Geography

null