TFDD: A High-Quality Image Dataset for Accurate Tomato Fruit Disease Detection and Classification
Description
Tomatoes are one of the most widely cultivated and economically significant crops worldwide. However, their productivity and quality are significantly affected by various diseases, including bacterial, viral, and fungal infections. Tomato Fruit Disease Detection (TFDD) dataset designed to support the development and evaluation of object detection models. The dataset contains 288 original images. The images were collected from a local tomato field in Bhashanchar, Faridpur, and a local market in Dhaka, Bangladesh . It consists of high-resolution images of tomatoes affected by different diseases, captured under natural field conditions such as Anthracnose, Blossom End Rot, Fruit Worm, Fruit Cracking, Late Blight, Mold, Early Blight, and Healthy. To enhance dataset diversity and improve model generalization, we applied several augmentation techniques, including horizontal and vertical flipping, rotations between -15° and +15°, brightness adjustments from -15% to +15%, and saturation variations between -20% and +20%. We augmented (3x) of the training images only. After augmentation we obtained 682 images (595 images for training, 58 images for validation, and 29 images for testing). Dataset Overview: Total Images: 682 Image Format: .jpg Image Size: 640x640 Annotation Format: .txt Number of Classes: 8 (Healthy and Diseased) Data Sources and Annotation: Field Location: Gafur Matubbor Dangi, Bhashanchar, Karirhat 7821, Sadarpur, Faridpur, Bangladesh Local Market Location: Khilkhet, Dhaka, Bangladesh. Captured Method: iPhone 13 and iPhone 15 Plus cameras Annotation Tool: Roboflow (Manual process) Annotation Guidelines: Bounding boxes Annotation Format (.txt): <class_id> <x_center> <y_center> <width> <height> Class Distributions: Class Name Raw Image count Anthracnose 27 Blossom_end_rot 31 Cracking 35 Early_Blight 20 Fruitworm 61 Healthy 27 Late_Blight 38 Mold 49
Files
Steps to reproduce
1. Download the TFDD zipped file from the dataset repository. 2. After unzipping the dataset, the directory structure will be organized as follows: TFDD/ ├── data.yaml # Dataset configuration for YOLO ├── summary.csv # Image–annotation summary for non-YOLO users ├── train/ │ ├── images/ # 595 JPEG images for training │ └── labels/ # 595 YOLO-format annotation files ├── valid/ │ ├── images/ # 58 JPEG images for validation │ └── labels/ # 58 YOLO-format annotation files └── test/ ├── images/ # 29 JPEG images for testing └── labels/ # 29 YOLO-format annotation files 3. No special software is required beyond a standard unzipping tool. To programmatically interact with the data, users may optionally use Python with libraries such as `pandas` and `os`.
Institutions
- American International University Bangladesh