Abstract: We show, for the first time, that neural networks trained only on synthetic data achieve state-of-the-art accuracy on the problem of 3D human pose and shape (HPS) estimation from real images ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results