DCGANGenerator

class deepinv.models.DCGANGenerator(output_size=64, nz=100, ngf=64, nc=3, dim=2)

Bases: Module

DCGAN Generator.

The DCGAN generator model was originally proposed by Radford et al.[1] and takes a latent sample as input.

Implementation taken from https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html

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

Parameters:

• output_size (int) – desired square size of output image. Choose from 64 or 128, defaults to 64

• nz (int) – latent dimension, defaults to 100

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

• nc (int) – number of image output channels, defaults to 3

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

---

References:

[1]

Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015.

forward(z, *args, **kwargs)

Generate an image

Parameters:

z (torch.Tensor) – latent vector

Examples using DCGANGenerator:

Imaging inverse problems with adversarial networks

Imaging inverse problems with adversarial networks