Deméter Rice Panicle Loss Detection Dataset

Description This dataset supports the development and evaluation of computer vision models for detecting rice grain loss in panicles. It was collected as part of the improvement of Deméter, a tool for automated rice loss estimation, developed at the Universidade Federal do Pampa (UNIPAMPA), Alegrete, RS, Brazil. Images were captured in a real rice field in the municipality of Alegrete, Rio Grande do Sul, Brazil, using a Samsung Galaxy S24 Ultra smartphone. Each original photograph was divided into four quadrants, yielding individual images at 4488 × 2072 pixels. Dataset Statistics Total images: 644 Total annotations: 59,454 Classes: `grain`, `loss` Annotation format: COCO (JSON) Annotation Annotations were produced manually and follow the [COCO format](https://cocodataset.org/#format-data), with bounding boxes for two classes representing individual rice grains (`grain`) and grains characterizing loss (`loss`). Intended Use This dataset is intended for training and evaluating object detection models, particularly in the context of small object detection in agricultural imagery.It was used to train a Faster R-CNN (ResNet-50 + FPN V2) model, achieving F2 = 0.8023 and Recall = 0.8037 for the loss class on the test set. Related Work This dataset was produced as part of an undergraduate thesis (TCC) at UNIPAMPA — Campus Alegrete, under the supervision of Prof. João Pablo Silva da Silva. License This dataset is released under the [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/) license.