Meta-Album AWA Micro: 800 Mammal Images for Few-Shot Learning

800 images of 20 mammal species, preprocessed to a uniform 128x128 pixel resolution. This micro version is derived from the Animals with Attributes 2 dataset, originally collected from public sources like Flickr in 2016 for transfer-learning benchmarks. The dataset was curated by Meta-Album for few-shot image classification tasks.

Use Cases

• Benchmarking few-shot image classification algorithms based on the 20 animal classes.
• Training attribute-based classification models based on the dataset's original design for attribute learning.
• Evaluating zero-shot learning performance on a standardized set of mammal images.
• Developing and testing image preprocessing pipelines for small-scale, square image datasets.

Strengths

• Uniform image size of 128x128 pixels after a described preprocessing pipeline involving cropping and resizing.
• Clearly defined scope with 20 classes and 800 total images, suitable for controlled experiments.
• Derived from a well-known benchmark dataset (Animals with Attributes 2) with a documented original purpose.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Image resolution is low (128x128 px) after preprocessing, which may limit detail for certain tasks.
• The dataset is a small subset (micro version) of the original source, which may not represent the full diversity of the parent dataset.

Provenance

Source

Animals with Attributes 2 (AWA2) dataset, collected from public sources like Flickr.

Collection Method

Images were collected from public sources in 2016, then cropped, squared, and resized to 128x128 pixels.

Time Range

Original images were collected in 2016.