r"""Subclass of nn.LayerNorm to handle fp16.

r""" Reduce feature maps into a single edge map

A residual block that can optionally change the number of channels. :param channels: the number of input channels. :param emb_channels

r"""NOTE: 1. Input image should has only one gray channel. 2. Input image size should be divisible by 8. 3. Sketch in the inpu

Transformer block for image-like data. First, project the input (aka embedding) and reshape to b, t, d. Then apply standard trans

A downsampling layer with an optional convolution. :param channels: channels in the inputs and outputs. :param use_conv: a bool determ

Transformer block for image-like data. First, project the input (aka embedding) and reshape to b, t, d. Then apply standard trans

An upsampling layer with an optional convolution. :param channels: channels in the inputs and outputs. :param use_conv: a bool determi

r""" Compact Dilation Convolution based Module

r""" Compact Spatial Attention Module

r"""DropPath but without rescaling and supports optional all-zero and/or all-keep.

Uses the OpenCLIP transformer encoder for text

Uses the OpenCLIP transformer encoder for text

Uses the OpenCLIP transformer encoder for text

Uses the OpenCLIP transformer encoder for text

r"""MiDaS v3.0 DPT-Large from ``https://github.com/isl-org/MiDaS''. Monocular depth estimation using dense prediction transformers.

AdaFactor pytorch implementation can be used as a drop in replacement for Adam original fairseq code: https://github.com/pytorch/fairseq/blob

r"""An infinite batch sampler.

r"""Differentiable all-gather.

r"""Differentiable all-reducd.

r"""Differentiable copy that reduces all gradients during backward.

r"""Differentiable scatter.

r""" CPDC, APDC can be converted to vanilla 3x3 convolution RPDC can be converted to vanilla 5x5 convolution

r"""Create a color palette (codebook) in the form of a 2D grid of colors. Further, the rightmost column has num_hues gradations from black to