Dataset: Exploring the Potential of Transfer Learning for Leaf Inclination Angle Distribution Assessment from Leveled Digital Photography

This dataset supports the manuscript "Exploring the Potential of Transfer Learning for Leaf Inclination Angle Distribution Assessment from Leveled Digital Photography". It contains leveled RGB photographs, leaf inclination angle distribution (LIAD) class labels, and associated species metadata. The dataset includes a model-development set with 2,271 images from 214 broadleaf trees and shrub species spanning 45 families, and an independent validation set with 441 images from 73 species. The five LIAD classes are planophile, plagiophile, uniform, spherical, and erectophile. The model-development dataset was used for training, validation, held-out testing, and ablation analysis of the LIAD transfer-learning configurations. The independent validation dataset was kept fully separate from model training, hyperparameter tuning, checkpoint selection, and ablation-based model selection. It was used exclusively for external validation and comparison with other models.