Educational OMR Answer Sheet Dataset for Automated Assessment via Object Detection

Overview This dataset contains scanned and photographed paper-based multiple-choice answer sheets designed for Optical Mark Recognition (OMR) using object detection models. The dataset was developed to support research on automated scoring frameworks based on deep learning object detection architectures. It has been used in the associated publication describing an automated paper-based multiple choice scoring framework. Two versions of the dataset are provided: datasetv1 – Version used in the original research paper. datasetv2 – Enhanced version with upgraded architecture (YOLOv11) and modularized detection tasks. All sensitive personal information has been anonymized using a reproducible masking procedure prior to release. Dataset Structure datasetv1 (Paper Version) This version corresponds to the dataset used in the published paper. Single unified detection model YOLO format annotations Standard split into: train validation test dataset.yaml defines class names and dataset paths classes.txt lists object categories This version supports end-to-end detection of all required OMR components within a single model. datasetv2 (Enhanced Version) This is the upgraded and modularized dataset version. Key improvements: Migrated to YOLOv11 Detection tasks separated into independent models Increased flexibility and improved maintainability The dataset is divided into three independent detection tasks: 1. data_answer Detection of multiple-choice answer bubbles. classes_answer.txt dataset_answer.yaml train / val / test split 2. data_info Detection of student information fields (e.g., ID regions, structured metadata areas). classes_info.txt dataset_info.yaml train / val / test split 3. data_marker Detection of alignment markers and reference anchors used for geometric correction. classes_marker.txt dataset_marker.yaml train / val / test split This modular design enables: Independent training Improved detection accuracy More scalable system design Easier experimentation and benchmarking Data Split Each dataset version provides: Training set Validation set Test set The split ensures fair evaluation and reproducibility. Anonymization All answer sheets were anonymized before publication. Personal identifiers such as names, signatures, and sensitive information have been masked using a reproducible automated masking procedure. Intended Use This dataset is intended for: Research on Optical Mark Recognition (OMR) Object detection benchmarking Educational assessment automation Computer vision research in document analysis License This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Citation If you use this dataset, please cite: The associated research article This dataset DOI