Instead of 2D key- point locations, [48,49] use 2D heatmaps [26,27] as input and learn convolutional neural networks for 3D pose regression. The aforementioned ...
確定! 回上一頁