FruitTraits v1.0: Fruit colour of fleshy-fruited woody plants measured in citizen-science photographs worldwide

FruitTraits v1.0 gives the colour of individual fleshy fruits of trees, shrubs and woody climbers measured in georeferenced iNaturalist photographs taken within the native ranges of the species and selected to spread over latitude, elevation, places and observers. In 81,368 photographs, 331,899 fruits were detected and outlined with an open-vocabulary detector (Grounding DINO) and a promptable segmentation model (Segment Anything 2.1), checked for outline quality, classified as single and ripe by classifiers trained on 1,367 rated fruits, and measured in CIELAB. The dataset gives the outline, colour and classification of every fruit and colour summaries for 34,911 photographs of 2,303 species from 154 families and 197 countries. Files: photos.csv.gz (one row per photograph), fruits.csv.gz (one row per outlined fruit), species.csv (one row per species), validation_photos.csv.gz (photographs of the technical validation), benchmark_fruits.csv and benchmark_rating_guide.md (rated benchmark fruits and the rules of the ratings), fruit_classifiers.zip (weights and cross-validated performance of the fruit classifiers), examples.zip and crops.zip (images of photographs licensed CC0 or CC BY, with fruit masks and credits), code.zip (code and settings files), data_dictionary.csv (description of every column of every table) and SHA256SUMS.txt (checksums of all files). README.md describes the files and how to select data. Tables are UTF-8 CSV files with a header row; missing values are empty fields, so that the ISO country code of Namibia (NA) is kept. Images in examples.zip and crops.zip keep the licences of their rights holders (CC0 or CC BY), and their reuse requires the credits given in examples_credits.csv and crops_credits.csv. Observations: iNaturalist research-grade observations shared through GBIF (doi:10.15468/dl.5evp9z). The FruitTrait R package (https://github.com/ehsanrahimi666/FruitTrait) reads the tables and recomputes fruit colour from the outlines.