Human Action Recognition Dataset for AutoML Training

An image dataset for training models to recognize human actions, curated by author beurkinger. It is hosted on Hugging Face and was last updated in July 2022. Platform tags indicate it contains between 10,000 and 100,000 image samples.

Use Cases

• Train an image classifier to predict action labels from visual frames.
• Benchmark model performance on human action recognition tasks using the provided image data.
• Fine-tune pre-trained vision models on a specific set of human actions for transfer learning.
• Develop multi-modal models if the dataset includes associated temporal or pose data inferred from the action recognition domain.

Strengths

• Contains an estimated 10,000 to 100,000 image samples based on platform size tags.
• Focuses on a well-defined computer vision task (human action recognition).
• Hosted on a reputable platform (Hugging Face) with versioning.

Limitations

• Exact number of rows, columns, and specific class distribution are unknown.
• Data may be temporally stale, with no updates since mid-2022.
• Geographic and demographic bias is likely present, with a 'us' region tag suggesting a U.S. focus.

Provenance

Source

Hugging Face dataset repository by author beurkinger.

Collection Method

Collection method is unspecified; likely aggregated or curated from existing sources for AutoML training.

Freshness

Last updated on 2022-07-18; no update frequency specified.

Geography

Platform tags suggest a primary focus on the United States ('Regionus').