Meta-Album Birds Dataset: 12,600 Images Across 315 Species

Meta-Album_BRD_Mini is a preprocessed subset of the Birds 400 - Species Image Classification dataset, originally created by Gerald Piosenka and curated by Dustin Carrion for Meta-Album. It contains 12,600 images across 315 bird species, each resized to 128x128 pixels. The dataset was released in March 2022 for use in few-shot image classification benchmarks.

Use Cases

• Fine-grained image classification based on bird species.
• Benchmarking few-shot learning algorithms on a natural image dataset.
• Studying model bias related to natural background contexts mentioned in the description.
• Analyzing class imbalance effects using the noted male-to-female image ratio.

Strengths

• Contains 12,600 images across 315 distinct classes, providing a substantial multi-class benchmark.
• Images are uniformly preprocessed to 128x128 pixels using an anti-aliasing filter, ensuring consistency.

Limitations

• The dataset is described as imbalanced regarding the ratio of male to female species images.
• Natural backgrounds may introduce bias, as some birds are frequently associated with specific environments like water.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Kaggle dataset '100-bird-species' by Gerald Piosenka.

Collection Method

Preprocessed (resized) from the original dataset for the Meta-Album benchmark collection.

Freshness

Created 01 March 2022.