blambert/ne_plant_photos

NE Plant Photos

Plant photographs from the iNaturalist open data archive, filtered to a New England bounding box (latitude 41 to 48, longitude -74 to -67), each labelled with the species it records and where it was observed.

This is the nature portion of blambert/ne_plant_classes - photos a vision language model judged to show a plant in the field, rather than a person, a microscope image, or something manmade - so photos of people and indoor shots are largely gone, but the labelling is model output and has not been checked against a hand-labelled sample.

114,054 images in a single train split, covering the species in the source archive at up to 100 photos per species.

Columns

• image - the photograph, embedded in the dataset
• file_name - the original filename, named after the iNaturalist photo id
• photo_id, photo_uuid, extension - iNaturalist photo identifiers
• license, observer_id - who took the photo and on what terms
• observation_uuid, observed_on, latitude, longitude - the observation the photo belongs to
• taxon_id, name, ancestry - the iNaturalist taxon id, its scientific name, and the /-separated ancestor taxon ids up to the root

Every row is a research-grade observation of an active species-rank taxon under Plantae, with the photo at least 750 pixels on each side.

Licensing

The photographs keep whichever license each iNaturalist observer chose, so the collection is not under a single license and most of it is non-commercial: CC-BY-NC, CC-BY, CC0, CC-BY-NC-SA and CC-BY-SA, with the exact terms per photo in the license column. Everything but the CC0 portion requires attribution.

Provenance

Built by plant-pics: the archive is filtered with filter.py, the images downloaded with download_pics.py, labelled by running vlm process with prompts/inat_classify.txt over them, and assembled here by build_dataset.py, which keeps the nature label and joins the filter's metadata back onto each image.