Meta-Album INS Extended: 170,506 Insect Images for Few-Shot Classification

170,506 preprocessed images of insects in natural habitats, collected by researchers and volunteers for the SPIPOLL citizen science project. The dataset contains 117 hierarchical classes and was curated by the National Museum of Natural History, Paris, with preprocessing by the Meta-Album project in March 2022. Images are cropped and resized to a uniform 128x128 pixel resolution.

Use Cases

• Training few-shot image classification models based on the 117 hierarchical insect classes.
• Benchmarking meta-learning algorithms on a standardized, preprocessed image dataset.
• Developing computer vision models for insect identification and biodiversity monitoring based on citizen-science imagery.
• Studying hierarchical classification techniques using the Order, Super-Family, Family, and Texa labels.

Strengths

• Large scale with 170,506 images, providing substantial data for model training.
• Images are preprocessed to a consistent 128x128 pixel size, reducing initial data preparation effort.
• Contains a detailed hierarchical label structure across 117 classes for nuanced classification tasks.
• Clear provenance from a major scientific institution (National Museum of Natural History, Paris) and a citizen science project.

Limitations

• Description metadata is limited; actual data quality and label consistency require manual inspection after download.
• License is CC BY-NC 2.0, which prohibits commercial use.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

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

Collection Method

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

Freshness

Created for Meta-Album on 01 March 2022; original data collection time range is unspecified.