Face Reconstruction From Deep Facial Embeddings Using A Convolutional Neural Network

in this paper, a subjective evaluation of static point clouds encoded with several codecs is described. Unlike other studies, a stereoscopic 3D display was used to visualize the 3D representation. A set of six point clouds were encoded using a set of state of the art point cloud coding solutions, notably the two MPEG codecs V-PCC and G-PCC, a deep learning solution RS-DLPCC that was the response to a call for evidence on point cloud coding of JPEG Pleno, and the popular DRACO codec. The results of this subjective quality evaluation using a 3D representation visualized in a stereoscopic display were compared with a previous subjective study that used the same content visualized in a 2D display. For that, the results of both tests were compared with the Pearson correlation, Spearman rank order correlation, the root mean square error and the outlier ratio. Moreover, the two subjective evaluation results were statistically analysed to seek for any statistical difference. The two subjective evaluations reveal a very high level of similarity.