Harmony4D: Multi-View Video of Close Human Contact Interactions

Harmony4D is a large-scale multi-view video dataset featuring in-the-wild close human–human contact interactions. It includes activities like wrestling, dancing, MMA, karate, fencing, and hugging, with dense ground-truth annotations for detection, tracking, and 3D pose estimation. The dataset was created by Voxel51 and last updated on March 18, 2026.

Use Cases

• Training detection and tracking models based on multi-view video of close-contact interactions.
• Developing 2D and 3D human pose estimation pipelines using the provided dense ground-truth annotations.
• Researching SMPL body mesh recovery for scenarios with physical interpenetration and occlusion.
• Benchmarking multi-person computer vision algorithms on in-the-wild close-contact activities like wrestling and dancing.

Strengths

• Focuses on close-contact scenarios where standard single-person pipelines are known to fail.
• Provides dense ground-truth annotations for multiple computer vision tasks.
• Covers a variety of in-the-wild interaction types, including wrestling, dancing, MMA, and hugging.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and dataset size are unknown, which may limit suitability assessment.
• License information is unknown, which may restrict usage.

Provenance

Source

Voxel51

Collection Method

Multi-view video collection of in-the-wild close human–human contact interactions.

Freshness

Last updated 2026-03-18 17:59:25; freshness should be verified.