Anthropometric measurements are essential for various industries, including healthcare, sports science, and fashion. Traditionally, these measurements are obtained manually or using specialized 3D scanners, methods that are often time-consuming, costly, and inaccessible. Estimating 3D body measurements from 2D images presents a significant challenge due to the ill-posed nature of inferring 3D geometry from 2D projections and the scarcity of large-scale datasets with accurate 3D annotations. To address these challenges, we propose AnthroFormer3D, a deep learning framework that automates the estimation of anthropometric measurements directly from 2D images. Our model leverages Vision Transformers (ViTs) to capture global context and spatial relationships critical for modeling human body shape. Additionally, we introduce a large-scale synthetic dataset comprising over 1.2 million images from 150,000 virtual subjects, designed to represent broad variations in height, weight, body mass index, and natural human poses. A key aspect of our method is the integration of auxiliary height and weight information into the model, enabling it to exploit complex non-linear correlations between these attributes and body shape. We explore early and late fusion strategies for incorporating these features into the ViT architecture. Experimental evaluations show that our models achieve superior accuracy compared to baseline methods, particularly when employing gender-specific training. The key contributions of this work are threefold: (1) the development of a Vision Transformer-based model specifically designed for anthropometric measurement estimation, (2) the creation of a large-scale, publicly available synthetic dataset to support training and evaluation, and (3) the design of an effective framework that integrates auxiliary attributes, such as height and weight, to improve predictive performance. Collectively, these contributions advance automated anthropometry and enable practical applications.