Channel-wise self-attention block.

Maintains (exponential) moving average of a set of parameters.

AFHQ Cat dataset.

InceptionC block patched for FID computation

InceptionA block patched for FID computation

copied modified from https://github.com/voletiv/self-attention-GAN-pytorch/blob/master/sagan_models.py#L29 copied modified from https://githu

CfgNode represents an internal node in the configuration tree. It's a simple dict-like container that allows for attribute-based access to ke

Pretrained InceptionV3 network returning feature maps

CelebA HQ dataset.

This class implements the D-Flow method for solving inverse problems, from the paper Ben-Hamu et al, "D-Flow: Differentiating through flows for contro

First InceptionE block patched for FID computation

Second InceptionE block patched for FID computation

r""" :math:`\ell_1` data fidelity term. In this case, the data fidelity term is defined as .. math:: f(x) = \|Ax-y\|_1.

r""" Base class for noise model. NoiseModel can be combined via :meth:`deepinv.physics.noise.NoiseModel.__mul__`, :param torch.Generato

Channel-wise self-attention block. Modified from DDPM.

Combine information from skip connections.

Conv2d layer with optimal upsampling and downsampling (StyleGAN2).

Linear layer with `default_init`.

Gaussian Fourier embeddings for noise levels.

NCSN++ model

NCSNv2 model architecture for 128px images.

NCSNv2 model architecture for 256px images.

The ResNet Blocks used in DDPM.

ResBlock adapted from DDPM.

AFHQ dataset.

CelebA HQ dataset.