Wildlife Scraped: Multi-Source Image Dataset Covering 122 Species Labels

A collection of 139,231 wildlife images across 122 species and sub-species labels, aggregated from multiple online sources. The dataset was created by Horama and last updated on 2026-04-29. Images are stored in Parquet shards compatible with HuggingFace's Image feature for direct decoding.

Use Cases

• Train image classification models based on the 122 species and sub-species labels.
• Benchmark multi-source data aggregation techniques based on images from iNaturalist, Wikimedia, Wikipedia, Flickr, and DuckDuckGo.
• Develop domain adaptation models based on the dataset's split into train, validation, and test subsets.
• Fine-tune pre-trained vision models for ecological monitoring based on the wildlife image content.

Strengths

• Contains 139,231 images, providing a substantial volume for model training.
• Covers 122 distinct labels, including species, sub-species, and domestic categories.
• Provides predefined splits: 97,456 training images (70%), 20,913 validation images (15%), and 20,862 test images (15%).
• Aggregates images from five public sources: iNaturalist, Wikimedia, Wikipedia, Flickr, and DuckDuckGo.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic, temporal, or source bias inherent to the listed platforms.

Provenance

Source

iNaturalist, Wikimedia, Wikipedia, Flickr, DuckDuckGo

Collection Method

Scraped from multiple online sources.

Freshness

Last updated 2026-04-29