On-tree mango instance segmentation dataset

Dataset created for on-tree mango fruit detection and segmentation as a part of mango fruit size estimation study. Image datasets were prepared for training of Convolutional Neural Network (CNN) based instance segmentation model and annotated using VGG Image Annotation tool (Dutta & Zisserman 2019) with polygon region annotation. Two folders contain cropped images and COCO style JSON annotation files and randomly separated into training and test image sets. Images were taken at nighttime with the use of artificial light (LED light panel), using Azure Kinect RGB-D camera and Basler ace acA2440-75uc RGB camera.

Datasets contain images from Honey Gold and Keitt mango cultivars and folders are organized as:

Folder 1 (individual-mango-snips) - contains tiled-images and annotation file - A total of 542 (train + test) tiled images of 640 x 540 pixels.

Folder 2 (tiled-images) - individual-mango-snips - Total 1200 (train + test) snips of variable dimensions (<150 pixels)

For anyone intended to use the dataset for their research purpose, please cite following related journal paper:

Neupane, C.; Koirala, A.; Walsh, K.B. In-Orchard Sizing of Mango Fruit: 1. Comparison of Machine Vision Based Methods for On-The-Go Estimation. Horticulturae 2022, 8, 1223. https://doi.org/10.3390/horticulturae8121223