GobhiSet: Dataset of manually and automatically annotated RGB images of early phenological stages of Brassica oleracea var. Botrytis

The dataset comprises a repository of manually and automatically annotated RGB images taken across multiple dates of Brassica oleracea, distributed in uniform crop spaces. It is designed to facilitate crop detection, segmentation and growth modelling based on manual and automatically annotated pixel information. The repository is publicly available here and consists of 132 original RGB images captured over 3 different dates. Each image has a dimension of 5472×3648 pixels. Manual annotations were performed using Visual Geometry Group Image Annotator (VIA) using bounding boxes and the results were saved in Common Objects in Context (COCO) segmentation format. The labelling information such as region and shape attributes are detailed in Javascript Object Notation (JSON). Other attributes such as individual crop ID and repetitiveness of individual crop specimens are described in the Comma Separated Values (CSV) version of the annotation. For creation of automated annotations, the manual annotations were trained over a framework of Grounding DINO + Segment Anything Model (SAM) and the labels were saved in Pascal Visual Object Classes (PASCAL VOC) format. The segmentation masks are provided as Portable Network Graphics (PNG) images. The automated annotations facilitate growth monitoring over the crop phenology through evaluation based on binary masks of individually identified crop specimens across different dates. The codes used for these processes are accessible to ensure transparency and reproducibility. The dataset not only provides annotation information but can also help in refinement of different machine learning models.