Meta-Album INS Mini: 4,110 Insect Images for Few-Shot Learning

France's National Museum of Natural History and the SPIPOLL citizen science project collected over 290,000 images of insects in natural habitats. This mini version contains 4,110 preprocessed images across 104 taxonomic classes, each resized to 128x128 pixels. The dataset was curated by Ihsan Ullah for the Meta-Album benchmark in March 2022.

Use Cases

• Benchmarking few-shot image classification models based on 104 insect classes.
• Training models for fine-grained visual categorization based on hierarchical insect taxonomy (Order, Super-Family, Family, Texa).
• Developing ecological monitoring tools using citizen-science sourced imagery of insects in natural environments.
• Studying domain adaptation for models trained on preprocessed, cropped, and resized natural images.

Strengths

• Contains 4,110 images, providing a substantial sample for few-shot learning tasks.
• Images are consistently preprocessed to a uniform 128x128 pixel size.
• Offers a hierarchical class structure with 104 distinct taxonomic labels.
• Sourced from a large, original collection of over 290,000 images validated by researchers and volunteers.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Images are cropped to squares and downscaled, which may remove contextual habitat information.

Provenance

Source

National Museum of Natural History, Paris and the SPIPOLL citizen science project.

Collection Method

Images were collected by researchers and volunteers via a website and mobile apps, then centrally stored and preprocessed (cropped and resized).

Geography

Primarily France, based on the source institutions.