Cats Vs Dogs: 10,000+ Images from the Microsoft Asirra Project

Microsoft provided this collection of 10,000 to 100,000 images of cats and dogs as a subset of the Asirra dataset. Originally used for a Kaggle competition, the data was designed to test computer vision capabilities against CAPTCHA-style image recognition tasks. This version has been cleaned to remove 1,738 corrupted images found in the original source.

Use Cases

• Training binary image classifiers to distinguish between 'cat' and 'dog' labels
• Benchmarking computer vision models on the Asirra CAPTCHA challenge subset
• Evaluating model performance on varied image resolutions and lighting conditions

Strengths

• Contains between 10,000 and 100,000 labeled images
• Crowdsourced annotations for binary classification
• Derived from the Microsoft Asirra CAPTCHA project

Limitations

• Limited to only two animal classes
• Potential for label noise inherent in crowdsourced CAPTCHA data
• Geographic and breed bias may exist based on the original Asirra collection source

Provenance

Source

Microsoft (Asirra dataset)

Collection Method

crowdsourced

Freshness

Last updated August 2024; however, the source material originates from a historical Kaggle competition.