medusa.preproc.align#

Module Contents#

medusa.preproc.align.estimate_alignment(v, topo, target=None, estimate_scale=False, device=DEVICE)[source]#

Aligment of a temporal series of 3D meshes to a target (which should have the same topology).

Parameters:

  • v (torch.tensor) – A float tensor with vertices of shape B (batch size) x V (vertices) x 3

  • topo (str) – Topology corresponding to v

  • target (torch.tensor) – Target to use for alignment; if None (default), a default template will be used

  • estimate_scale (bool) – Whether the alignment may also involve scaling

  • device (str) – Either ‘cuda’ (GPU) or ‘cpu’

Returns:

mat – A float tensor with affine matrices of shape B (batch size) x 4 x 4

Return type:

torch.tensor