DIBaS: Digital Images of Bacterial Species for Few-Shot Classification

Meta-Album BCT Micro is a preprocessed version of the Digital Images of Bacteria Species (DIBaS) dataset, containing 800 images across 20 bacterial classes. The dataset was created by Romain Mussard for the Meta-Album benchmark in March 2022, with original images sourced from researchers including Bartosz Zielinski and Krzysztof Misztal. Images were split and resized to 128x128 pixels and processed with blob normalization to reduce color bias.

Use Cases

• Benchmarking few-shot learning algorithms based on the standardized 20-class, 800-image structure.
• Training bacterial colony classifiers based on preprocessed 128x128 microscopic images.
• Studying bias-aware model performance based on the dataset's blob normalization to reduce color importance.
• Developing computer vision models for microscopic image analysis based on the domain-specific imagery.

Strengths

• Contains 800 preprocessed images, providing at least 108 samples per class for more balanced few-shot learning.
• Images are uniformly sized at 128x128 pixels and processed with blob normalization, offering a consistent input format.
• Clear provenance with original author citations and a dedicated Meta-Album benchmark page for documentation.

Limitations

• Row count and file format details for the underlying data files are unknown, which may limit suitability assessment.
• The dataset is a derived, preprocessed subset; the original image context and full resolution are lost.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Digital Images of Bacteria Species (DIBaS) dataset, preprocessed for the Meta-Album benchmark.

Collection Method

Original microscopic images were split into smaller patches and resized.

Time Range

Original data associated with a 2017 publication; preprocessed version created in 2022.

Freshness

Created on 01 March 2022; last update date is unknown.

Geography

null