A Balanced Papaya Leaf Disease Dataset via CycleGAN-based Generative Augmentation

This dataset provides a balanced collection of 7,000 high-resolution images across seven categories of papaya leaf diseases, including Bacterial Blight, Carica Insect Hole, Curled Yellow Spot, Healthy Leaf, Mosaic Virus, Pathogen Symptoms, and Yellow Necrotic Spots Holes. The raw data was meticulously collected from agricultural fields in Savar, Ashulia, and Daffodil Smart City, Bangladesh, between January and October 2024 to capture seasonal variations. To ensure high ground-truth reliability, all images were independently validated by senior agronomists through a majority voting consensus. To overcome the severe class imbalance found in the original 1,684 images, we utilized Cycle-Consistent Generative Adversarial Networks (CycleGAN) to synthesize high-fidelity samples for minority classes. The final dataset is perfectly balanced with 1,000 images per class, providing a robust foundation for training unbiased deep learning models for automated plant pathology and agricultural computer vision tasks.