Pose2Mesh: Graph Convolutional Network for 3D Human Pose and Mesh Recovery from a 2D Human Pose

Most of the recent deep learning-based 3D human pose and mesh estimation\nmethods regress the pose and shape parameters of human mesh models, such as\nSMPL and MANO, from an input image. The first weakness of these methods is an\nappearance domain gap problem, due to different image appearance between train\ndata from controlled environments, such as a laboratory, and test data from\nin-the-wild environments. The second weakness is that the estimation of the\npose parameters is quite challenging owing to the representation issues of 3D\nrotations. To overcome the above weaknesses, we propose Pose2Mesh, a novel\ngraph convolutional neural network (GraphCNN)-based system that estimates the\n3D coordinates of human mesh vertices directly from the 2D human pose. The 2D\nhuman pose as input provides essential human body articulation information,\nwhile having a relatively homogeneous geometric property between the two\ndomains. Also, the proposed system avoids the representation issues, while\nfully exploiting the mesh topology using a GraphCNN in a coarse-to-fine manner.\nWe show that our Pose2Mesh outperforms the previous 3D human pose and mesh\nestimation methods on various benchmark datasets. For the codes, see\nhttps://github.com/hongsukchoi/Pose2Mesh_RELEASE.\n