Synthetic Human Motion Video and Depth Data

8 million synthetic images and 75 hours of video data depict human motion. The dataset contains 27,480 image sequences at 30fps and 1280x720 resolution, created by Intelligent-Systems. It was last updated in March 2026.

Use Cases

• Train human pose estimation models using the camera/scene ground truth CSV+JSON files.
• Develop video-based action recognition algorithms on the 75 hours of MP4 video sequences.
• Benchmark monocular depth estimation models using the associated BEDLAM2-depth data.
• Generate synthetic training data for models requiring precise 3D scene and camera parameters.

Strengths

• 8 million images provide a large-scale synthetic data source.
• 75 hours of video at 30fps offers substantial temporal data.
• Includes camera and scene ground truth (CSV+JSON) for all sequences.

Limitations

• Data is entirely synthetic, which may not fully capture real-world noise and variations.
• The total size is large (11TB for images, 160GB for video), requiring significant storage.

Provenance

Source

Intelligent-Systems, Max Planck Institute for Intelligent Systems.

Collection Method

Synthetically generated video and image data.

Freshness

Last updated in March 2026.