Meta_Album_SPT_Mini: 73 Sports Action Images for Few-Shot Learning

2,920 images across 73 sports categories, resized to 128x128 pixels for few-shot learning benchmarks. The dataset is a processed version of the 100-Sports collection, originally gathered from internet searches and deduplicated. It was released under a public domain license by Meta-Album in March 2022.

Use Cases

• Benchmarking few-shot image classification models based on the 73 distinct sports categories.
• Training meta-learning algorithms on a domain of human actions using the provided image collection.
• Evaluating model generalization on sports imagery from internet sources after preprocessing steps like resizing.

Strengths

• Contains 2,920 images, providing a defined scale for model training and evaluation.
• Covers 73 distinct sports classes, offering a multi-class classification challenge.
• Images were processed to a uniform 128x128 pixel size and duplicate images were removed.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• The original source is internet imagery, which may reflect biases in representation or image quality.

Provenance

Source

Processed from the '100 Sports Image Classification' dataset by Gerald Piosenka on Kaggle.

Collection Method

Images were gathered from internet searches, deduplicated, and resized by Meta-Album.