AnthroVision Dataset

This research provides a novel public dataset for automated nutritional status assessment in children. The dataset contains multi-angle images and anthropometric measurements collected from clinical (All India Institute of Medical Sciences (AIIMS) Jodhpur) and community settings (spanning 5 government schools in the district) in Rajasthan, India. The dataset contains samples labeled with anthropometric measurements for regression and classification labels based on World Health Organization (WHO) and Centers for Disease Control and Prevention (CDC) standards.

Dataset Details

The dataset is a comprehensive collection of data points, each represented by a tuple (Ii, yri, yci), where: Ii is the set of multi-angle images for the i-th subject, encompassing both full-body and facial views. yri is the vector of anthropometric measurements for the i-th subject, including age, height, weight, Body Mass Index (BMI), Mid-Upper Arm Circumference (MUAC), and Head Circumference (HC), serving as regression labels. Additionally, weight-for-age, height-for-age, and BMI-for-age z-scores have been calculated based on WHO and CDC growth standards to create classification labels. yci is classified into categories such as healthy/unhealthy based on BMI z-score(-2>BMIz>1), underweight or wasting (inferred from age wise MUAC cutoffs and weight-for-age < -2 SD), and overall category (based on presence of stunting/underweight/wasting i.e., height-for-age Z-score (HAZ) < -2SD or agewise MUAC cutoffs or weight-for-age < -2 SD). Adhering to WHO and CDC standards ensures the dataset's validity and alignment with clinical practices in nutritional status assessment.

Dataset Composition

The dataset comprises 2,141 subjects (1,326 in clinical and 815 in community settings). Each subject contributes approximately eight multi-pose angles totalling 16,938 images. The dataset was developed in collaboration with government institutions, hospitals, and schools, ensuring ethical compliance.