mmagic.models.editors.stylegan2.ada.augment

Module Contents

Classes

AugmentPipe

Augmentation pipeline include multiple geometric and color

Functions

matrix(*rows[, device])	Constructing transformation matrices.
translate2d(tx, ty, **kwargs)	Construct 2d translation matrix.
translate3d(tx, ty, tz, **kwargs)	Construct 3d translation matrix.
scale2d(sx, sy, **kwargs)	Construct 2d scaling matrix.
scale3d(sx, sy, sz, **kwargs)	Construct 3d scaling matrix.
rotate2d(theta, **kwargs)	Construct 2d rotating matrix.
rotate3d(v, theta, **kwargs)	Constructing 3d rotating matrix.
translate2d_inv(tx, ty, **kwargs)	Construct inverse matrix of 2d translation matrix.
scale2d_inv(sx, sy, **kwargs)	Construct inverse matrix of 2d scaling matrix.
rotate2d_inv(theta, **kwargs)	Construct inverse matrix of 2d rotating matrix.

Attributes

conv2d
wavelets

mmagic.models.editors.stylegan2.ada.augment.conv2d

mmagic.models.editors.stylegan2.ada.augment.wavelets

mmagic.models.editors.stylegan2.ada.augment.matrix(*rows, device=None)

Constructing transformation matrices.

Parameters

device (str|torch.device, optional) – Matrix device. Defaults to None.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.translate2d(tx, ty, **kwargs)

Construct 2d translation matrix.

Parameters

• tx (float) – X-direction translation amount.

• ty (float) – Y-direction translation amount.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.translate3d(tx, ty, tz, **kwargs)

Construct 3d translation matrix.

Parameters

• tx (float) – X-direction translation amount.

• ty (float) – Y-direction translation amount.

• tz (float) – Z-direction translation amount.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.scale2d(sx, sy, **kwargs)

Construct 2d scaling matrix.

Parameters

• sx (float) – X-direction scaling coefficient.

• sy (float) – Y-direction scaling coefficient.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.scale3d(sx, sy, sz, **kwargs)

Construct 3d scaling matrix.

Parameters

• sx (float) – X-direction scaling coefficient.

• sy (float) – Y-direction scaling coefficient.

• sz (float) – Z-direction scaling coefficient.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.rotate2d(theta, **kwargs)

Construct 2d rotating matrix.

Parameters

theta (float) – Counter-clock wise rotation angle.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.rotate3d(v, theta, **kwargs)

Constructing 3d rotating matrix.

Parameters

• v (torch.Tensor) – Luma axis.

• theta (float) – Rotate theta counter-clock wise with v as the axis.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.translate2d_inv(tx, ty, **kwargs)

Construct inverse matrix of 2d translation matrix.

Parameters

• tx (float) – X-direction translation amount.

• ty (float) – Y-direction translation amount.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.scale2d_inv(sx, sy, **kwargs)

Construct inverse matrix of 2d scaling matrix.

Parameters

• sx (float) – X-direction scaling coefficient.

• sy (float) – Y-direction scaling coefficient.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

mmagic.models.editors.stylegan2.ada.augment.rotate2d_inv(theta, **kwargs)

Construct inverse matrix of 2d rotating matrix.

Parameters

theta (float) – Counter-clock wise rotation angle.

Returns

Transformation matrices in np.ndarry or torch.Tensor

format.

Return type

ndarry | Tensor

class mmagic.models.editors.stylegan2.ada.augment.AugmentPipe(xflip=0, rotate90=0, xint=0, xint_max=0.125, scale=0, rotate=0, aniso=0, xfrac=0, scale_std=0.2, rotate_max=1, aniso_std=0.2, xfrac_std=0.125, brightness=0, contrast=0, lumaflip=0, hue=0, saturation=0, brightness_std=0.2, contrast_std=0.5, hue_max=1, saturation_std=1, imgfilter=0, imgfilter_bands=[1, 1, 1, 1], imgfilter_std=1, noise=0, cutout=0, noise_std=0.1, cutout_size=0.5)

Bases: torch.nn.Module

Augmentation pipeline include multiple geometric and color transformations.

Note: The meaning of arguments are written in the comments of __init__ function.

forward(images, debug_percentile=None)