Ecuadorian native and common fruit image dataset for embedded classification (eNFCG-Frutas)

This dataset contains 1,295 RGB images across eight classes, developed to support research on embedded fruit classification systems using convolutional neural networks deployed on low-power GPU platforms (NVIDIA Jetson Nano). The dataset comprises four Ecuadorian native fruits (sweet granadilla/grenadilla, yellow pitahaya/dragonfruit, tree tomato/tamarillo, and papaya), three common fruits (apple, banana, and tomato), and one background class. Directory structure and class distribution Class Train Val Test Total apple 105 33 15 153 background 106 43 21 170 banana 106 32 15 153 dragonfruit 110 35 20 165 grenadilla 110 35 20 165 papaya 110 35 20 165 tamarillo 110 40 21 171 tomato 107 30 16 153 Total 864 283 148 1,295 Class-to-index mapping From labels.txt, in alphabetical order, matching the model's output index order: apple background banana dragonfruit grenadilla papaya tamarillo tomato Acquisition All fruit samples were commercially acquired from local markets; the study did not involve wild plant collection, protected species sampling, or field harvesting. Images were captured using the Data Capture Control interface of the jetson-inference framework, connected to a camera on the NVIDIA Jetson Nano platform. Filenames follow a YYYYMMDD-HHMMSS.jpg convention reflecting capture date and time; class and split membership are encoded in the folder structure, not the filename.