A CNN trained to recognize the human shape PCA parameters. Example inputs are included in data file. The output of the network is a 20 by 1 dimension vector in which each numbers corresponds to PCA eigenvectors weights. For details of the method, please check the paper "HS-Nets : Estimating Human Body Shape from Silhouettes with Convolutional Neural Networks". For data generation, please contact me.