PatchGANDiscriminator

class deepinv.models.PatchGANDiscriminator(input_nc=3, ndf=64, n_layers=3, use_sigmoid=False, batch_norm=True, bias=True, dim=2)

Bases: Module

PatchGAN Discriminator model.

This discriminator model was originally proposed by Isola et al.[1] and classifies whether each patch of an image is real or fake.

Implementation adapted from Kupyn et al.[2].

See Imaging inverse problems with adversarial networks for how to use this for adversarial training.

Parameters:

• input_nc (int) – number of input channels, defaults to 3

• ndf (int) – hidden layer size, defaults to 64

• n_layers (int) – number of hidden conv layers, defaults to 3

• use_sigmoid (bool) – use sigmoid activation at end, defaults to False

• batch_norm (bool) – whether to use batch norm layers, defaults to True

• bias (bool) – whether to use bias in conv layers, defaults to True

• dim (str, int) – Whether to build 2D or 3D network (if str, can be “2”, “2d”, “3D”, etc.)

---

References:

[1]

Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, 1125–1134. 2017.

[2]

Orest Kupyn, Volodymyr Budzan, Mykola Mykhailych, Dmytro Mishkin, and Jiří Matas. Deblurgan: blind motion deblurring using conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, 8183–8192. 2018.

forward(x)

Forward pass of discriminator model.

Parameters:

x (torch.Tensor) – input image

Examples using PatchGANDiscriminator:

Imaging inverse problems with adversarial networks

Imaging inverse problems with adversarial networks