iNaturalist (iNat)

Data
Image data
Biology
Ecology
Computer vision
Multimodal learning
Benchmarking
Citizen science
CLIP
Zero-shot learning
Author

Chris Endemann

Published

April 3, 2025

About this resource

iNaturalist (iNat) is one of the largest community-contributed biodiversity datasets, with millions of photos of plants, animals, and fungi submitted by citizen scientists around the world. These images are accompanied by rich metadata — including time, location, and species ID — and are verified through community consensus.

The iNat2024 (iNat24) subset, used in benchmarks like INQUIRE, includes over 5 million images and provides a real-world testbed for evaluating species classification, visual reasoning, and multimodal retrieval tasks.

What makes iNat valuable for AI?

Most large-scale image datasets are curated or scraped from the internet. In contrast, iNat data reflects real-world ecological observations: images are often messy, off-center, and diverse in setting, but rich in scientific value. This makes iNat an ideal source for developing and testing robust machine learning models.

Researchers use iNat to:

  • Train and evaluate models on fine-grained species classification
  • Benchmark retrieval tasks involving complex visual and contextual cues
  • Explore zero-shot generalization using CLIP or other vision-language models
  • Support conservation efforts and ecological research at scale

Example uses

iNat images have powered major ML benchmarks and tools, including the INQUIRE benchmark, which uses iNat24 to evaluate retrieval models on expert-level ecological queries

Access and attribution

iNat data is accessible through the iNaturalist API, the Global Biodiversity Information Facility (GBIF), and various competition archives. Data licensing follows Creative Commons guidelines, and attribution to individual observers is required.

Questions?

If you’re using iNat or want to ask about its structure or use cases, feel free to post in the ML+X Nexus Q&A forum.

See also