Meta-Album DOG Mini: 4,800 Images of 120 Dog Breeds for Fine-Grained Categorization

Stanford Dogs Dataset is a collection of over 20,000 images of 120 dog breeds sourced from ImageNet for fine-grained image categorization. The Meta-Album DOG Mini version contains 4,800 images preprocessed to 128x128 pixels by cropping and resizing. Researchers Aditya Khosla, Nityananda Jayadevaprakash, Bangpeng Yao, and Li Fei-Fei created the original dataset, which was released in 2011 and later adapted by Meta-Album in March 2022.

Use Cases

• Train fine-grained image classification models based on 120 distinct dog breeds.
• Benchmark few-shot learning algorithms using the standardized 128x128 pixel image format.
• Study intra-class variation in visual data based on differences in color, pose, and occlusion mentioned in the description.
• Evaluate model robustness to background variation based on images taken in man-made environments.

Strengths

• Contains 4,800 images across 120 distinct classes, providing a multi-class classification challenge.
• Images are preprocessed to a consistent 128x128 pixel resolution, simplifying input pipelines.
• Derived from the well-known Stanford Dogs Dataset, which has over 20,000 original images and is cited in computer vision literature.

Limitations

• The number of images per class and original resolution were not balanced, which may introduce bias.
• Description metadata is limited; actual data quality requires manual inspection after download.
• Images have been cropped and resized, which may remove contextual information present in the original frames.

Provenance

Source

Stanford Dogs Dataset (http://vision.stanford.edu/aditya86/ImageNetDogs/), sourced from ImageNet.

Collection Method

Images were cropped to squares and resized to 128x128 pixels using an anti-aliasing filter for the Meta-Album release.

Freshness

The Meta-Album version was created on 01 March 2022; the original dataset dates from 2011.

Geography

Dogs breeds are from worldwide sources, but specific geography is not detailed.