Meta-Album INS2: 800 Images of 20 Insect Pest Classes for Few-Shot Learning

800 square-cropped images of insect pests across 20 classes, preprocessed to 128x128 pixels for few-shot learning benchmarks. This dataset is a curated subset of the larger IP102 benchmark, created by Ihsan Ullah for the Meta-Album project in March 2022. The original IP102 dataset contains over 75,000 images across 102 categories with a hierarchical taxonomy linking pests to specific agricultural products.

Use Cases

• Few-shot image classification based on 20 distinct insect pest classes.
• Benchmarking meta-learning algorithms using the standardized 128x128 image format.
• Training lightweight vision models for agricultural pest detection based on preprocessed, square-cropped images.
• Studying hierarchical classification based on the dataset's described taxonomy linking pests to agricultural products.

Strengths

• Images are uniformly preprocessed to a square 128x128 pixel format, ensuring consistency.
• Provides a clear, manageable scale with 800 images across 20 classes for few-shot learning tasks.
• Derived from a larger, cited benchmark (IP102) containing over 75,000 images, suggesting a foundation in prior research.

Limitations

• Description metadata is limited; actual data quality and potential label noise require manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's specific geographic and temporal coverage is not detailed, which may limit understanding of its representativeness.

Provenance

Source

IP102: A Large-Scale Benchmark Dataset for Insect Pest Recognition (https://github.com/xpwu95/IP102)

Collection Method

Images were cropped into perfect squares and resized from the original IP102 dataset.