A comprehensive, high-resolution benchmark repository containing 9,567 labeled RGB photographs across 20 distinct mango cultivars. Photographed in Naogaon, Bangladesh with separate Fruit and Leaf domain partitions and predefined leak-free group splits for computer vision and agricultural AI.
Current Deposit Status: Pending Mendeley Data Approval
The dataset files, metadata manifests, and descriptions have been submitted to Mendeley Data and are in curator review. Download archives will become directly accessible once approved and published.
Lakh_Fazli is present only in the fruit subset (not in leaves). Evaluate Fruit and Leaf domains independently.
Total Images
9,567
Distinct Varieties
20 Cultivars
Split Discipline
Zero Leak
Collection Region
Naogaon, BD
Predefined splits carefully grouped by source-groups to eliminate training and evaluation data leakage.
6,497 Images across 20 Varieties
4,548
~70.0%977
~15.0%972
~15.0%Fruit grouping in the manifest relies on visual source groups. All images standardized to RGB 224 × 224 pixels.
3,070 Images across 19 Varieties
2,243
~73.1%419
~13.6%408
~13.3%
Reflects group-based partition without cross-split source group leakage. Notice: Lakh_Fazli excluded.
Displays image count per variety across Train, Validation, and Test splits
Search, sort, and inspect class distribution across both Fruit and Leaf partitions.
| Variety Label | Domains Available | Fruit Train | Fruit Val | Fruit Test | Fruit Total | Leaf Train | Leaf Val | Leaf Test | Leaf Total | Combined Total |
|---|---|---|---|---|---|---|---|---|---|---|
* Note: Lakh_Fazli is unavailable in the leaf collection subset.
Explore repository hierarchical structure and column annotations for metadata CSV manifests.
Mango Fruit and Leaf Variety Image Dataset/
├── README.md
├── README.txt
├── UPLOAD_DESCRIPTION.txt
├── metadata/
│ ├── contributors.csv
│ └── source_groups.csv
├── fruit/
│ ├── metadata.csv
│ └── images/{train,val,test}/<variety>/*.jpg
└── leaf/
├── metadata.csv
└── images/{train,val,test}/<variety>/*.jpg
image_path: Relative to respective subset folder (fruit/ or leaf/).
sha256: Available in fruit manifests; optional in leaf rows.
| Field | Type | Description & Values |
|---|---|---|
| image_path | string | Relative to subset directory (e.g. images/train/Amrapali/img_001.jpg). |
| label | string | Mango variety class label matching exact folder nomenclature. |
| split | enum | train, val, or test. |
| source_group_id | string/id | Collection grouping key, mapped directly to metadata/source_groups.csv. |
| latitude, longitude | float | Coordinates in decimal degrees recorded for the source group centroid. |
| geographic_description | string | Recorded location narrative (Naogaon, Rajshahi, Bangladesh). |
| image_width, image_height | integer | Standardized processed image resolution: 224 × 224 pixels. |
| sha256 | hex/string | Cryptographic image hash. May be blank for select leaf images. |
⚠️ Geographic Notice: Coordinates represent source group collection sites, not individual photo GPS tags. No source-group ID is shared between train, validation, and test splits.
Load Mango Fruit and Leaf subsets effortlessly with standard deep learning frameworks.
Designed for supervised image classification, transfer learning (ResNet, EfficientNet, Vision Transformers), domain adaptation between fruit & leaf morphological phenotypes, and automated precision agriculture.
Cultivar sample counts are inherently imbalanced (e.g., BARI_4 has 627 fruits vs Gaurmati with 112). Authors strongly encourage reporting Macro-F1, Precision, and Recall alongside top-1 accuracy.
Deposited for publication on Mendeley Data under CC BY 4.0 license.
The dataset submission has completed field curation and formatting. The Mendeley Data editorial team is reviewing the deposit. Direct download buttons below will automatically link to the permanent files once the review concludes and the DOI is minted.
Full 6,497 fruit images with metadata.csv and train/val/test subdirectories.
Full 3,070 leaf images with metadata.csv and partitioned splits.
Includes both Fruit and Leaf subsets, full metadata tables, source-group coordinates, and contributor records.
If you utilize the Mango Fruit and Leaf Variety Image Dataset in your academic research, benchmarks, or agricultural machine learning publications, please cite the published Mendeley Data repository.
License: CC BY 4.0 (Creative Commons Attribution 4.0 International)
Repository Deposit: Mendeley Data
Contact: Author contact emails listed in metadata/contributors.csv
@dataset{mango_fruit_leaf_variety_2026,
title = {Mango Fruit and Leaf Variety Image Dataset},
author = {Al Galib, Arshad and Abbas, Jakariya and Bashir, Abdul and
Uddowla, Md. Anis and Islam, Md. Aminul and Seyam, Redwan Shiddiki},
year = {2026},
publisher = {Mendeley Data (Under Review)},
address = {Naogaon, Rajshahi, Bangladesh},
note = {9,567 RGB Images (224x224), 20 Varieties, Fruit and Leaf domains. DOI pending Mendeley approval},
license = {Creative Commons Attribution 4.0 International (CC BY 4.0)}
}