MMHAR-28: Human Action Recognition Across Four Sensing Modalities

MMHAR-28 is a multimodal human action recognition dataset containing 28 action classes. The data was collected in two sessions, with Session 1 focusing on single-person sports and exercise actions and Session 2 on two-person interaction activities. It was authored by 'issai' and last updated on the Hugging Face platform in March 2026.

Use Cases

• Train action classification models based on the four described sensing modalities: RGB, Depth, Thermal, and Event.
• Benchmark multimodal fusion techniques for human action recognition based on the dataset's structure.
• Study the differences between single-person and two-person interaction activities based on the two-session collection design.

Strengths

• Covers 28 distinct human action classes.
• Provides data across four different sensing modalities: RGB, Depth, Thermal, and Event.
• Data collection is structured into two distinct sessions for single-person and two-person activities.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

issai

Collection Method

Likely collected via sensors capturing RGB, depth, thermal, and event data.

Time Range

null

Freshness

Last updated 2026-03-26 05:28:06; freshness should be verified.

Geography

null