Meta-Album PLK Extended: 473,000 Plankton Images for Classification

Meta_Album_PLK_Extended contains 473,273 preprocessed plankton images across 102 classes, sourced from the Woods Hole Oceanographic Institution. The dataset was created by researchers including Heidi M. Sosik and preprocessed by Ihsan Ullah for the Meta-Album benchmark in March 2022. Images are standardized to 128x128 pixels with a blurred background to isolate plankton specimens.

Use Cases

• Training image classification models to identify plankton species based on the 102 classes.
• Benchmarking few-shot learning algorithms using the standardized image format and class structure.
• Developing models for automated plankton monitoring and ecological analysis based on the controlled imaging environment described.
• Evaluating preprocessing techniques for biological imaging, given the described background subtraction and resizing methodology.

Strengths

• Large scale with 473,273 images, providing substantial data for model training.
• High class diversity with 102 distinct plankton categories.
• Standardized image preprocessing ensures consistent 128x128 pixel dimensions.
• Clear provenance from the Woods Hole Oceanographic Institution, a leading research center.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The preprocessing method involving background blurring may alter or obscure some original image features.
• Data may reflect temporal and geographic bias inherent to the original collection source and method.

Provenance

Source

Woods Hole Oceanographic Institution (WHOI), via the WHOI-Plankton GitHub repository.

Collection Method

Collected using an Imaging FlowCytobot (IFCB), with images preprocessed by creating a squared background, applying a Gaussian blur, and resizing.

Freshness

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