MPII Human Pose Dataset: 1,160 Images of 410 Human Activities

A preprocessed subset of the MPII Human Pose benchmark containing 1,160 images of people performing 410 distinct activities, with each image labeled for activity classification. The dataset was created by Mykhaylo Andriluka and colleagues at the Max Planck Institute for Informatics and released in 2014. All images have been resized to 128x128 pixels as part of the Meta-Album collection for few-shot learning.

Use Cases

• Benchmarking few-shot image classification models based on the 410 activity labels.
• Training models for human action recognition using the labeled activity categories.
• Developing pose estimation techniques on preprocessed, uniformly sized images.
• Evaluating meta-learning algorithms on a standardized human action dataset.

Strengths

• Includes 1,160 images across 29 classes, providing a focused subset for few-shot learning.
• Images are systematically collected from YouTube videos covering 410 human activities.
• All images are preprocessed to a uniform 128x128 pixel resolution.
• Derived from a well-known, state-of-the-art benchmark (MPII Human Pose) with a clear citation.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• The mini version contains significantly fewer images than the original benchmark of ~25,000 images.
• Image resolution is reduced to 128x128 pixels, which may limit detail for fine-grained pose analysis.

Provenance

Source

MPII Human Pose Dataset (Max Planck Institute for Informatics)

Collection Method

Images were systematically collected from YouTube videos using a taxonomy of human activities.