xnn.cnn.models.steerable.rotate_fields#

xnn.cnn.models.steerable.rotate_fields(x, R, fields)[source]#

Rotate a stack of fields by a symmetry of the cube: grid and fibers.

Applies rotate_voxels() to every channel and D^l(R) to the components of each field of order l (the induced representation of eq 1, for a proper rotation R).

Parameters:

x (Tensor) – Fields of shape (B, field_dim(fields), s, s, s).
R (Tensor) – A proper rotation among the signed permutation matrices.
fields (sequence of int) – Multiplicities of the stack.

Returns:

The rotated stack, same shape.

Return type:

Tensor