Tomato_detection

This dataset comprises 1,600 tomato images collected for binary classification of Fresh and Rotten tomatoes, intended for use in agricultural quality control and post-harvest inspection systems. The original dataset consisted of 800 images, of which 200 images were reserved for validation and 200 for testing to ensure unbiased evaluation. The remaining 400 training images were augmented using Roboflow to generate 1,200 training samples, resulting in a 75/13/13 train-validation-test split. Each image was pre-processed with auto-orientation (EXIF stripping) and resized to 640×640 pixels using a fit-with-black-edges strategy to maintain aspect ratio. To improve model generalization, the following augmentations were applied to produce 3 versions of each training image: vertical flip, random rotation between −15° and +15°, saturation adjustment between −15% and +15%, exposure adjustment between −10% and +10%, and Gaussian blur of up to 0.5 pixels. The dataset is annotated in YOLOv8 bounding box format with two classes — Fresh and Rotten — and is publicly available under a CC BY 4.0 license via Roboflow Universe.