Human-Animal-Cartoon: Multimodal Video Clips for Domain Adaptation

A collection of 3,381 video clips depicting seven actions performed by humans, animals, and cartoon figures. The dataset includes three modalities: video, audio, and pre-computed optical flow. It was created by author hdong51 and last updated on the platform in April 2026.

Use Cases

• Train action recognition models based on the seven labeled actions.
• Research domain adaptation techniques based on the three distinct visual domains (human, animal, cartoon).
• Develop multimodal models based on the provided video, audio, and optical flow data.
• Benchmark cross-domain generalization performance for video understanding tasks.

Strengths

• Contains 3,381 video clips, providing a substantial base for training.
• Includes three data modalities (video, audio, optical flow) per clip.
• Covers three distinct visual domains with approximately 1,000 clips each.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect source bias inherent to being collected from the internet.

Provenance

Source

Collected from the internet by author hdong51.

Collection Method

Video clips were collected from the internet.

Freshness

Last updated 2026-04-30 09:30:17; freshness should be verified.