VABD-IFMS v1.0: A Multi-Person Scene Dataset for Visual Acuity Behavior Detection

This work presents the design, development, and validation of the Visual Acuity Behavior Dataset (VABD-IFMS), a novel image dataset for studying behaviors associated with visual strain in realistic, multi-person educational settings. Addressing the lack of public resources for training robust computer vision models for visual acuity screening, this project executed a rigorous data collection protocol approved by an institutional Research Ethics Committee (Brazil, CAAE: 83336124.7.0000.5162). Methodological Approach and Dataset Structure: Unlike single-label classification datasets, VABD-IFMS is designed to reflect the complexity of real-world classroom environments. Each image (frame) in the dataset captures a scene containing multiple individuals (typically 6 to 13 subjects per frame), who may simultaneously exhibit a variety of different behaviors. The annotations, provided in the standard JSON format as exported by Label Studio, contain bounding boxes for each person identified and a corresponding class label from 7 predefined behaviors. This structure is specifically intended for the development and validation of object detection models (e.g., YOLO, SSD) that can be connected to live camera systems. Data Availability and Access Conditions: To protect participant privacy and in accordance with the project's ethical protocol, the dataset files (images and annotations) are not publicly hosted but are available to researchers under a controlled access protocol. This metadata record serves to ensure the dataset is Findable, Citable, and Reusable under defined conditions. Researchers interested in accessing the dataset must: Submit a formal request to the corresponding author ([e-mail]) detailing the intended research project and purpose of use. Provide evidence of ethical approval for their project from their own institution's Research Ethics Committee (or equivalent). Sign a Data Use Agreement (DUA) committing to uphold the same standards of data security, confidentiality, and purpose limitation outlined in this work, and agreeing not to redistribute the data. The documentation, including the full final report and the annotation manual, is provided with this record. The dataset itself consists of over 7,000 images, available in both processed (224x224, CLAHE) and raw formats, with [Seu número final] high-quality annotations. VABD-IFMS is a fundamental tool that enables the development and evaluation of new AI-driven solutions for the early identification of students who may require professional ophthalmological assessment.