Rotate#

class deepinv.transform.Rotate(*args, limits: float = 360.0, multiples: float = 1.0, positive: bool = False, interpolation_mode: InterpolationMode = InterpolationMode.NEAREST, **kwargs)[source]#

Bases: Transform

2D Rotations.

Generates n_trans randomly rotated versions of 2D images with zero padding (without replacement).

Picks integer angles between -limits and limits, by default -360 to 360. Set positive=True to clip to positive degrees. For exact pixel rotations (0, 90, 180, 270 etc.), set multiples=90.

By default, output will be cropped/padded to input shape. Set constant_shape=False to let output shape differ from input shape.

See deepinv.transform.Transform for further details and examples.

Parameters:

  • limits (float) – images are rotated in the range of angles (-limits, limits).

  • multiples (float) – angles are selected uniformly from \(\pm\) multiples of multiples. Default to 1 (i.e integers) When multiples is a multiple of 90, no interpolation is performed.

  • positive (bool) – if True, only consider positive angles.

  • n_trans (int) – number of transformed versions generated per input image.

  • rng (torch.Generator) – random number generator, if None, use torch.Generator(), defaults to None

Examples using Rotate:#

Image transforms for equivariance & augmentations

Image transforms for equivariance & augmentations

Image transformations for Equivariant Imaging

Image transformations for Equivariant Imaging

Self-supervised learning with Equivariant Imaging for MRI.

Self-supervised learning with Equivariant Imaging for MRI.