Dataset for: Image recognition for native potatoes

Raw or collected database consisting of 3417 bounding boxes containing the following varieties: Peruanita (1015), Amarilla del Centro (1017), Huayro Rojo (166), and Huamantanga (1019). These were labeled using the image labeling program, generating labels for each variety (1 for Peruanita, 2 for Amarilla del Centro, 3 for Huayro Rojo, and 4 for Huamantanga). The label codes consist of four digits: the first two digits represent the location of the bounding box relative to the generated central quadrant by YOLO, and the last two digits represent the location of the tuber relative to the center within that quadrant (e.g., 3, 0.28, 0.54, 0.13, 0.17). After data collection, data preparation is performed to verify the quality of the collected data. The data has been curated one by one using the Roboflow software to regenerate the bounding boxes. This involved addressing issues such as blurry images, overlapping of the same potato type, overlapping of different potato types, overlapping with other tubers, and scenarios with other objects, resulting in 3430 bounding boxes. To improve the data quality, data augmentation techniques were applied, which randomly modify the rotation, noise, brightness, mosaics, and saturation of the images. As a result of this improvement process, the total number of bounding boxes increased to 10290, tripling its initial size. This final dataset is referred to as the "Data Augmented" dataset.