BioDex

Bird species image classification dataset: 55 species, ~28,752 images, built by downloading observation photos from GBIF and iNaturalist, supplemented with a small number of images from the Birds 525 Species Kaggle dataset for a few underrepresented species.

Dataset structure

• imagefolder format: one directory per species, directory name = class label.
• Single train split. Train/val/test splits are not defined yet — held back until data augmentation is finalized.

Collection process

Candidate images were pulled from GBIF and iNaturalist media URLs, then filtered through a bird detector (YOLO) and a species classifier confidence check. Images that failed detection or had low classification confidence were discarded before being added here.

License / why this repo is gated

Each image comes from an individual GBIF/iNaturalist observation and carries whatever license the original observer/photographer set (e.g. CC0, CC-BY, CC-BY-NC) — this has not been tracked or verified per image yet. Until that's sorted out, the repo stays gated rather than openly redistributed.

Usage

from datasets import load_dataset

ds = load_dataset("santianwandter/biodex")