HA-Multi-Samples v2: Multimodal Human Activity Data with Blurred Video

A multimodal human activity dataset organized per-episode and per-modality. It contains face-blurred RGB video, raw stereo pair images, full-hand tactile sensing, and distributed body IMU data captured during everyday household tasks. The dataset was created by humanarchive and released on Hugging Face, with a last update recorded on 2026-05-21.

Use Cases

• Train multimodal activity recognition models based on synchronized video, tactile, and IMU data.
• Develop privacy-aware computer vision algorithms based on face-blurred RGB video.
• Research sensor fusion techniques based on the combination of visual, inertial, and tactile modalities.
• Benchmark models for fine-grained activity analysis based on data from everyday household tasks.

Strengths

• Dataset provides synchronized data from multiple modalities including video, stereo images, tactile sensing, and IMUs.
• The episode set, frame counts, durations, and task labels are bit-identical to a previous v1 release, ensuring consistency for existing analyses.
• Video data has been processed with face-blurring (Meta EgoBlur Gen2), addressing privacy concerns.

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

humanarchive

Collection Method

Captured during everyday household tasks.

Freshness

Last updated 2026-05-21 08:59:33; freshness should be verified.