This is an adaptation of the DISCONet[1] paper(which estimates hand pose) for human 3D pose estimation from 2D pose
Thanks to weigq for the shared pytorch dataloaders and Human3.6M 2D to 3D datasets.
References
[1] D. Bouchacourt, M. P. Kumar, S. Nowozin, ”DISCO Nets: DISsimilarity COefficient Networks”, NIPS2016