Meta-Album Boats (Micro): 800 Images of 20 Maritime Vessel Classes

800 square images of 20 different maritime vessel classes, each resized to 128x128 pixels. This dataset is a preprocessed micro version of the larger MARVEL dataset, created by the Meta-Album project for few-shot learning benchmarks. The original data was authored by Gundogdu et al. and released in 2017, with the Meta-Album version created in March 2022.

Use Cases

• Benchmarking few-shot image classification models based on the 20 distinct boat classes.
• Training lightweight computer vision models on the 128x128 pixel image format.
• Developing domain adaptation techniques for vehicle recognition using the maritime vessel domain.
• Conducting educational experiments in computer vision with a curated, smaller-scale dataset.

Strengths

• Contains 800 images across 20 distinct classes, providing a defined scope for few-shot learning.
• All images are preprocessed to a consistent 128x128 pixel square format.
• Derived from the larger MARVEL dataset, which contains over 138,000 images, suggesting a substantive original source.

Limitations

• The micro version contains only 800 images, which may limit model training for non-few-shot tasks.
• Specific image counts per class and the original variable resolutions are not detailed for this subset.
• The description does not specify the exact preprocessing steps for each image beyond resizing and squaring.

Provenance

Source

MARVEL: A Large-Scale Image Dataset for Maritime Vessels (Gundogdu et al., 2017).

Collection Method

Images were preprocessed by duplicating rows or columns to create squares, then resized using an anti-aliasing filter.