Prediction of Anthropometric Data Based on Ladder Network
Haocan Xu, Jituo Li, Jiaman Li, Guodong Lu · 2019
The inadequacy of training data is an acute contradiction in anthropometric data prediction. To address this problem, a novel anthropometric data prediction method based on Ladder network is proposed. On one hand, some volunteers in various pose are captured with RGB cameras, whose anthropometric data (e.g., chest circumference) are measured manually. On the other hand, a series of unlabeled sample data are collected through virtual human bodies. After segmenting human body silhouettes from the background, a semi-supervised network (e.g., Ladder network) can be trained to predict the ratio between anthropometric data and height. Therefore, anthropometric data of arbitrary human body can be inferred with height, which can be potentially used in 3D human modeling and clothing industry. Experiments show that our method can achieve acceptable results with limited training data.