layerspp.py 🚧

Source: LION/models/diffusion/NCSNpp_helpers/layerspp.py

Warning

This file has not yet received a complete narrative and docstring audit. Its public source-level API is listed automatically below.

Layers for defining NCSN++.

conv1x1

conv3x3

NIN

default_init

class GaussianFourierProjection(embedding_size=256, scale=1.0)

Gaussian Fourier embeddings for noise levels.

GaussianFourierProjection.forward(self, x)

No docstring is available.

class Combine(dim1, dim2, method='cat')

Combine information from skip connections.

Combine.forward(self, x, y)

No docstring is available.

class AttnBlockpp(channels, skip_rescale=False, init_scale=0.0, qkv_init_scale=0.1, force_fp32_attention=False)

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

AttnBlockpp.forward(self, x)

No docstring is available.

class Upsample(in_ch=None, out_ch=None, with_conv=False, fir=False, fir_kernel=(1, 3, 3, 1))

No docstring is available.

Upsample.forward(self, x)

No docstring is available.

class Downsample(in_ch=None, out_ch=None, with_conv=False, fir=False, fir_kernel=(1, 3, 3, 1))

No docstring is available.

Downsample.forward(self, x)

No docstring is available.

class ResnetBlockDDPMpp(act, in_ch, out_ch=None, temb_dim=None, conv_shortcut=False, dropout=0.1, skip_rescale=False, init_scale=0.0, temb_activation=True)

ResBlock adapted from DDPM.

ResnetBlockDDPMpp.forward(self, x, temb=None)

No docstring is available.

class ResnetBlockBigGANpp(act, in_ch, out_ch=None, temb_dim=None, up=False, down=False, dropout=0.1, fir=False, fir_kernel=(1, 3, 3, 1), skip_rescale=True, init_scale=0.0, temb_activation=True)

No docstring is available.

ResnetBlockBigGANpp.forward(self, x, temb=None)

No docstring is available.