Meta-Album DOG: 800 Images of 20 Dog Breeds for Fine-Grained Categorization

Meta_Album_DOG_Micro is a preprocessed subset of the Stanford Dogs Dataset, containing 800 square images across 20 dog breeds. The original dataset was created by researchers from Stanford University for fine-grained image categorization, with images sourced from ImageNet. This micro version was created by Dustin Carrion in March 2022, with images cropped and resized to 128x128 pixels.

Use Cases

• Fine-grained image classification based on 20 dog breed categories.
• Model training and evaluation for tasks with high intra-class variation due to pose, color, and occlusion.
• Benchmarking computer vision models on a standardized, preprocessed image dataset.

Strengths

• Contains 800 images across 20 distinct dog breed classes.
• Images are preprocessed to a uniform 128x128 pixel resolution.
• Derived from the well-known Stanford Dogs Dataset, which contains over 20,000 images from 120 breeds.

Limitations

• Dataset size is limited to 800 images, which may be small for some deep learning applications.
• Images have been cropped and resized, which may alter original composition and detail.
• The description notes significant background variation and class imbalance in the original source, which may persist.

Provenance

Source

Stanford Dogs Dataset (http://vision.stanford.edu/aditya86/ImageNetDogs/)

Collection Method

Images and annotations were sourced from ImageNet and preprocessed by cropping to squares and resizing.