An Annotated Image Dataset of Mango, Lychee, and Jackfruit Leaf Diseases from the Chittagong Hill Tracts of Bangladesh
Description
This dataset contains 15,062 smartphone-captured RGB leaf images of three economically important fruit crops of Bangladesh: mango (Mangifera indica), lychee (Litchi chinensis), and jackfruit (Artocarpus heterophyllus). Images were collected from government-managed gardens and additional field sites in Rangamati Hill District, within the Chittagong Hill Tracts of Bangladesh, between October 2025 and May 2026. The final validated release contains 7,609 mango images, 5,445 lychee images, and 2,008 jackfruit images. It contains 39 crop-condition classes: 11 mango, 11 lychee, and 17 jackfruit classes, covering healthy leaves, visible disease symptoms, insect and pest damage, physiological conditions, and physical damage. Jackfruit (2,008 images, 17 classes): Algal Leaf Spot (231), Cercospora Leaf Spot (174), Dried Leaf (3), Healthy Leaf (501), Insect Damage (365), Leaf Blight (78), Leaf Chlorosis (38), Leaf Distortion (23), Leaf Roller Caterpillar (4), Leaf Spot (178), Leaf Spot & Senescence (37), Mealybug Infestation (66), Mechanical Damage (21), Powdery Mildew (39), Scale Insect Infestation (8), Sooty Mold (201), and Whitefly Infestation (41). Lychee (5,445 images, 11 classes): Anthracnose (555), Dieback (380), Healthy Leaves (786), Insect Damage (1,739), Leaf Blight (90), Leaf Blight & Anthracnose (76), Leaf Curl & Erinose Mite (685), Leaf Curl & Sooty Mold (54), Leaf Spot (575), Red Rust (159), and Sooty Mold (346). Mango (7,609 images, 11 classes): Anthracnose (2,675), Bacterial Leaf Spot (709), Dieback (315), Fungal (32), Healthy (954), Insect Damage (1,795), Mango Malformation (11), Mite Infestation (3), Powdery Mildew (389), Red Rust (249), and Sooty Mold (477). Images were captured in natural orchard and field conditions using several Android smartphones, and retain variation in illumination, background, orientation, symptom severity, and resolution. Images were cropped to retain relevant leaf regions and symptoms; no synthetic augmentation, background replacement, colour correction, denoising, filtering, or artificial enhancement was applied. Four contributors participated in annotation, and final labels were reviewed by an agriculture officer. Labels are expert-reviewed visual field assessments, not laboratory-confirmed pathogen identifications. Before Version 3 release, all 15,062 final .jpg files passed full pixel decoding. Of 1,087 files identified during quality control, 1,085 recoverable files were re-encoded and two unreadable files were removed. Initial exact-duplicate analysis identified 78 groups; no exact duplicate groups remained after cleanup. The archive contains species- and class-organised folders, metadata.csv, a metadata script, and documentation. Metadata fields are filename, species, class label, resolution, capture date when available, file size, and relative path. Complete image-level device identifiers and camera-wise counts were not recorded; device-balanced analysis is not directly supported.
Files
Steps to reproduce
1. Download the Version 3 archive file, All_Fruits_Dataset_V3.zip. 2. Extract the archive into a local directory. 3. The extracted All_Fruits directory contains Mango, Lychee, and Jackfruit class folders. 4. Use metadata.csv for programmatic image access and class labels. 5. Each metadata record corresponds to one image file and includes filename, species, class_label, resolution, capture_date when available, file_size_kb, and relative_path. 6. The final Version 3 release contains 15,062 JPEG-format images in 39 classes. 7. The generate_metadata_csv.py script can be used to regenerate metadata.csv from the released folder hierarchy.
Institutions
- Rangamati Science and Technology UniversityChittagong, Chittagong