Yongho Jeong

In this study, we propose a solution for automatically measuring body circumferences by utilizing the built-in LiDAR sensor in mobile devices. Traditional body measurement methods mainly rely on 2D images or manual measurements. This research, however, utilizes 3D depth information to enhance both accuracy and efficiency. By employing HRNet-based keypoint detection and transfer learning through deep learning, the precise locations of body parts are identified and combined with depth maps to automatically calculate body circumferences. Experimental results demonstrate that the proposed method exhibits a relative error of up to 8% for major body parts such as waist, chest, hip, and buttock circumferences, with waist and buttock measurements recording low error rates below 4%. Although some models showed error rates of 7.8% and 7.4% in hip circumference measurements, this was attributed to the complexity of 3D structures and the challenges in selecting keypoint locations. Additionally, the use of depth map-based keypoint correction and regression analysis significantly improved accuracy compared to conventional 2D-based measurement methods. The real-time processing speed was also excellent, ensuring stable performance across various body types.

In this study, we propose a solution for automatically measuring body circumferences by utilizing the built-in LiDAR sensor in mobile devices. While traditional body measurement methods primarily rely on 2D images or manual measurements, this research leverages 3D depth information to enable more accurate and efficient measurements. By employing HRNet-based keypoint detection and transfer learning through deep learning, the precise locations of body parts are identified and combined with depth maps to automatically calculate body circumferences. Experimental results demonstrate that the proposed method exhibits a relative error of up to 8\% for major body parts such as waist, chest, hip, and buttock circumferences, with waist and buttock measurements recording low error rates below 4\%. Although some models showed error rates of 7.8\% and 7.4\% in hip circumference measurements, this was attributed to the complexity of 3D structures and the challenges in selecting keypoint locations. Additionally, the use of depth map-based keypoint correction and regression analysis significantly improved accuracy compared to conventional 2D-based measurement methods. The real-time processing speed was also excellent, ensuring stable performance across various body types.

The key to semi-supervised semantic segmentation is to assign the appropriate pseudo-label to the pixels of unlabeled images. Recently, various approaches to consistency-based training and the filtering of reliable pseudo-labels have shown remarkable results. Nonetheless, there are still issues to be addressed. We find that recent approaches have specific problems in common. In pseudo-labels for training unlabeled images, we confirm that false foreground class pseudo-labels are mostly caused by background class confusion, not confusion between different foreground classes. To solve this problem, we propose a foreground and background discrimination model for semi-supervised semantic segmentation. Our proposed model is trained using a novel approach called multi-view integrated ensemble (MVIE) via output perturbation. Experimental results in various partition protocols show that our approach outperforms the existing state of the art (SOTA) in binary prediction on unlabeled data, and the segmentation model trained with the help of our model outperforms existing models.