3D anthropometric measurements have a variety of applications in industrial design and architecture (e.g. vehicle seating and cockpits), Clothing (e.g. military uniforms), Ergonomics (e.g. seating) and Medicine (e.g. nutrition and diabetes) etc. Therefore, there is a need for systems that can accurately extract human body measurements. Current methods estimate human body measurements from 3D scans, resulting in a heavy data collection burden. Moreover, minor variations in camera angle, distance, and body postures may significantly affect the measurement accuracy. In response to these challenges, this paper introduces a focused human body model for accurately extracting anthropometric measurements. Concretely, we design a Bypass Network based on CNN and ResNet architectures, which augments the frozen backbone SMPLer-X with additional feature extraction capabilities. On the other hand, to boost the efficiency of training a large-scale model, we integrate a dynamical loss function that automatically recalibrates the weights to make the network focus on targeted anthropometric parts. In addition, we construct a multimodal body measurement benchmark dataset consisting of depth, point clouds, mesh and corresponding body measurements to support model evaluation and future anthropometric measurement research. Extensive experiments on both open-source and the proposed human body datasets demonstrate the superiority of our approach over existing counterparts, including the current mainstream commercial body measurement software.