Meta-Album Boats: 138,367 Images of 26 Maritime Vessel Classes

Meta-Album's Boats dataset contains 138,367 square images of 26 different maritime vessel classes, resized to 128x128 pixels. It is a preprocessed version of the MARVEL dataset, created by Erhan Gundogdu et al. and released as part of the Meta-Album collection in March 2022. The images were processed by duplicating borders to create squares without cropping, as boats occupy most of the frame.

Use Cases

• Few-shot image classification based on 26 distinct maritime vessel classes.
• Training and evaluating computer vision models on preprocessed, square image data.
• Benchmarking model performance on a dataset with over 138,000 images.
• Research into domain adaptation for vehicle recognition tasks.

Strengths

• Contains 138,367 images, providing a substantial volume of data for model training.
• Includes 26 distinct classes of maritime vessels, enabling multi-class classification.
• Images are preprocessed to a uniform 128x128 pixel size, simplifying input pipelines.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is released under a CC BY-NC 4.0 license, restricting commercial use.
• The preprocessing method involved duplicating image borders, which may introduce artifacts.

Provenance

Source

MARVEL: A Large-Scale Image Dataset for Maritime Vessels (Gundogdu et al., ACCV 2016).

Collection Method

Images were collected and annotated for maritime vessels, then preprocessed by the Meta-Album team.