You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
860 lines
31 KiB
Python
860 lines
31 KiB
Python
"""Classes for generative image modeling"""
|
|
|
|
import logging
|
|
import math
|
|
from abc import abstractmethod
|
|
from typing import Iterable, List, Optional, Tuple, Union
|
|
|
|
import torch as th
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
from einops import rearrange
|
|
from torch.utils.checkpoint import checkpoint
|
|
|
|
from imaginairy.modules.attention import SpatialTransformer
|
|
from imaginairy.modules.sgm.diffusionmodules.util import (
|
|
avg_pool_nd,
|
|
conv_nd,
|
|
linear,
|
|
normalization,
|
|
timestep_embedding,
|
|
zero_module,
|
|
)
|
|
from imaginairy.modules.sgm.video_attention import SpatialVideoTransformer
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def exists(val):
|
|
return val is not None
|
|
|
|
|
|
class AttentionPool2d(nn.Module):
|
|
"""
|
|
Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
spacial_dim: int,
|
|
embed_dim: int,
|
|
num_heads_channels: int,
|
|
output_dim: Optional[int] = None,
|
|
):
|
|
super().__init__()
|
|
self.positional_embedding = nn.Parameter(
|
|
th.randn(embed_dim, spacial_dim**2 + 1) / embed_dim**0.5
|
|
)
|
|
self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
|
|
self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
|
|
self.num_heads = embed_dim // num_heads_channels
|
|
self.attention = QKVAttention(self.num_heads)
|
|
|
|
def forward(self, x: th.Tensor) -> th.Tensor:
|
|
b, c, _ = x.shape
|
|
x = x.reshape(b, c, -1)
|
|
x = th.cat([x.mean(dim=-1, keepdim=True), x], dim=-1)
|
|
x = x + self.positional_embedding[None, :, :].to(x.dtype)
|
|
x = self.qkv_proj(x)
|
|
x = self.attention(x)
|
|
x = self.c_proj(x)
|
|
return x[:, :, 0]
|
|
|
|
|
|
class TimestepBlock(nn.Module):
|
|
"""
|
|
Any module where forward() takes timestep embeddings as a second argument.
|
|
"""
|
|
|
|
@abstractmethod
|
|
def forward(self, x: th.Tensor, emb: th.Tensor):
|
|
"""
|
|
Apply the module to `x` given `emb` timestep embeddings.
|
|
"""
|
|
|
|
|
|
class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
|
|
"""
|
|
A sequential module that passes timestep embeddings to the children that
|
|
support it as an extra input.
|
|
"""
|
|
|
|
def forward(
|
|
self,
|
|
x: th.Tensor,
|
|
emb: th.Tensor,
|
|
context: Optional[th.Tensor] = None,
|
|
image_only_indicator: Optional[th.Tensor] = None,
|
|
time_context: Optional[int] = None,
|
|
num_video_frames: Optional[int] = None,
|
|
):
|
|
from imaginairy.modules.sgm.diffusionmodules.video_model import VideoResBlock
|
|
|
|
for layer in self:
|
|
module = layer
|
|
|
|
if isinstance(module, TimestepBlock) and not isinstance(
|
|
module, VideoResBlock
|
|
):
|
|
x = layer(x, emb)
|
|
elif isinstance(module, VideoResBlock):
|
|
x = layer(x, emb, num_video_frames, image_only_indicator)
|
|
elif isinstance(module, SpatialVideoTransformer):
|
|
x = layer(
|
|
x,
|
|
context,
|
|
time_context,
|
|
num_video_frames,
|
|
image_only_indicator,
|
|
)
|
|
elif isinstance(module, SpatialTransformer):
|
|
x = layer(x, context)
|
|
else:
|
|
x = layer(x)
|
|
return x
|
|
|
|
|
|
class Upsample(nn.Module):
|
|
"""
|
|
An upsampling layer with an optional convolution.
|
|
:param channels: channels in the inputs and outputs.
|
|
:param use_conv: a bool determining if a convolution is applied.
|
|
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
|
|
upsampling occurs in the inner-two dimensions.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
channels: int,
|
|
use_conv: bool,
|
|
dims: int = 2,
|
|
out_channels: Optional[int] = None,
|
|
padding: int = 1,
|
|
third_up: bool = False,
|
|
kernel_size: int = 3,
|
|
scale_factor: int = 2,
|
|
):
|
|
super().__init__()
|
|
self.channels = channels
|
|
self.out_channels = out_channels or channels
|
|
self.use_conv = use_conv
|
|
self.dims = dims
|
|
self.third_up = third_up
|
|
self.scale_factor = scale_factor
|
|
if use_conv:
|
|
self.conv = conv_nd(
|
|
dims, self.channels, self.out_channels, kernel_size, padding=padding
|
|
)
|
|
|
|
def forward(self, x: th.Tensor) -> th.Tensor:
|
|
assert x.shape[1] == self.channels
|
|
|
|
if self.dims == 3:
|
|
t_factor = 1 if not self.third_up else self.scale_factor
|
|
x = F.interpolate(
|
|
x,
|
|
(
|
|
t_factor * x.shape[2],
|
|
x.shape[3] * self.scale_factor,
|
|
x.shape[4] * self.scale_factor,
|
|
),
|
|
mode="nearest",
|
|
)
|
|
else:
|
|
x = F.interpolate(x, scale_factor=self.scale_factor, mode="nearest")
|
|
if self.use_conv:
|
|
x = self.conv(x)
|
|
return x
|
|
|
|
|
|
class Downsample(nn.Module):
|
|
"""
|
|
A downsampling layer with an optional convolution.
|
|
:param channels: channels in the inputs and outputs.
|
|
:param use_conv: a bool determining if a convolution is applied.
|
|
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
|
|
downsampling occurs in the inner-two dimensions.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
channels: int,
|
|
use_conv: bool,
|
|
dims: int = 2,
|
|
out_channels: Optional[int] = None,
|
|
padding: int = 1,
|
|
third_down: bool = False,
|
|
):
|
|
super().__init__()
|
|
self.channels = channels
|
|
self.out_channels = out_channels or channels
|
|
self.use_conv = use_conv
|
|
self.dims = dims
|
|
stride = 2 if dims != 3 else ((1, 2, 2) if not third_down else (2, 2, 2))
|
|
if use_conv:
|
|
# logger.info(f"Building a Downsample layer with {dims} dims.")
|
|
# logger.info(
|
|
# f" --> settings are: \n in-chn: {self.channels}, out-chn: {self.out_channels}, "
|
|
# f"kernel-size: 3, stride: {stride}, padding: {padding}"
|
|
# )
|
|
# if dims == 3:
|
|
# logger.info(f" --> Downsampling third axis (time): {third_down}")
|
|
self.op = conv_nd(
|
|
dims,
|
|
self.channels,
|
|
self.out_channels,
|
|
3,
|
|
stride=stride,
|
|
padding=padding,
|
|
)
|
|
else:
|
|
assert self.channels == self.out_channels
|
|
self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
|
|
|
|
def forward(self, x: th.Tensor) -> th.Tensor:
|
|
assert x.shape[1] == self.channels
|
|
|
|
return self.op(x)
|
|
|
|
|
|
class ResBlock(TimestepBlock):
|
|
"""
|
|
A residual block that can optionally change the number of channels.
|
|
:param channels: the number of input channels.
|
|
:param emb_channels: the number of timestep embedding channels.
|
|
:param dropout: the rate of dropout.
|
|
:param out_channels: if specified, the number of out channels.
|
|
:param use_conv: if True and out_channels is specified, use a spatial
|
|
convolution instead of a smaller 1x1 convolution to change the
|
|
channels in the skip connection.
|
|
:param dims: determines if the signal is 1D, 2D, or 3D.
|
|
:param use_checkpoint: if True, use gradient checkpointing on this module.
|
|
:param up: if True, use this block for upsampling.
|
|
:param down: if True, use this block for downsampling.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
channels: int,
|
|
emb_channels: int,
|
|
dropout: float,
|
|
out_channels: Optional[int] = None,
|
|
use_conv: bool = False,
|
|
use_scale_shift_norm: bool = False,
|
|
dims: int = 2,
|
|
use_checkpoint: bool = False,
|
|
up: bool = False,
|
|
down: bool = False,
|
|
kernel_size: int = 3,
|
|
exchange_temb_dims: bool = False,
|
|
skip_t_emb: bool = False,
|
|
):
|
|
super().__init__()
|
|
self.channels = channels
|
|
self.emb_channels = emb_channels
|
|
self.dropout = dropout
|
|
self.out_channels = out_channels or channels
|
|
self.use_conv = use_conv
|
|
self.use_checkpoint = use_checkpoint
|
|
self.use_scale_shift_norm = use_scale_shift_norm
|
|
self.exchange_temb_dims = exchange_temb_dims
|
|
|
|
if isinstance(kernel_size, Iterable):
|
|
padding = [k // 2 for k in kernel_size]
|
|
else:
|
|
padding = kernel_size // 2
|
|
|
|
self.in_layers = nn.Sequential(
|
|
normalization(channels),
|
|
nn.SiLU(),
|
|
conv_nd(dims, channels, self.out_channels, kernel_size, padding=padding),
|
|
)
|
|
|
|
self.updown = up or down
|
|
|
|
if up:
|
|
self.h_upd = Upsample(channels, False, dims)
|
|
self.x_upd = Upsample(channels, False, dims)
|
|
elif down:
|
|
self.h_upd = Downsample(channels, False, dims)
|
|
self.x_upd = Downsample(channels, False, dims)
|
|
else:
|
|
self.h_upd = self.x_upd = nn.Identity()
|
|
|
|
self.skip_t_emb = skip_t_emb
|
|
self.emb_out_channels = (
|
|
2 * self.out_channels if use_scale_shift_norm else self.out_channels
|
|
)
|
|
if self.skip_t_emb:
|
|
# logger.info(f"Skipping timestep embedding in {self.__class__.__name__}")
|
|
assert not self.use_scale_shift_norm
|
|
self.emb_layers = None
|
|
self.exchange_temb_dims = False
|
|
else:
|
|
self.emb_layers = nn.Sequential(
|
|
nn.SiLU(),
|
|
linear(
|
|
emb_channels,
|
|
self.emb_out_channels,
|
|
),
|
|
)
|
|
|
|
self.out_layers = nn.Sequential(
|
|
normalization(self.out_channels),
|
|
nn.SiLU(),
|
|
nn.Dropout(p=dropout),
|
|
zero_module(
|
|
conv_nd(
|
|
dims,
|
|
self.out_channels,
|
|
self.out_channels,
|
|
kernel_size,
|
|
padding=padding,
|
|
)
|
|
),
|
|
)
|
|
|
|
if self.out_channels == channels:
|
|
self.skip_connection = nn.Identity()
|
|
elif use_conv:
|
|
self.skip_connection = conv_nd(
|
|
dims, channels, self.out_channels, kernel_size, padding=padding
|
|
)
|
|
else:
|
|
self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
|
|
|
|
def forward(self, x: th.Tensor, emb: th.Tensor) -> th.Tensor:
|
|
"""
|
|
Apply the block to a Tensor, conditioned on a timestep embedding.
|
|
:param x: an [N x C x ...] Tensor of features.
|
|
:param emb: an [N x emb_channels] Tensor of timestep embeddings.
|
|
:return: an [N x C x ...] Tensor of outputs.
|
|
"""
|
|
if self.use_checkpoint:
|
|
return checkpoint(self._forward, x, emb)
|
|
else:
|
|
return self._forward(x, emb)
|
|
|
|
def _forward(self, x: th.Tensor, emb: th.Tensor) -> th.Tensor:
|
|
if self.updown:
|
|
in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
|
|
h = in_rest(x)
|
|
h = self.h_upd(h)
|
|
x = self.x_upd(x)
|
|
h = in_conv(h)
|
|
else:
|
|
h = self.in_layers(x)
|
|
|
|
if self.skip_t_emb:
|
|
emb_out = th.zeros_like(h)
|
|
else:
|
|
emb_out = self.emb_layers(emb).type(h.dtype)
|
|
while len(emb_out.shape) < len(h.shape):
|
|
emb_out = emb_out[..., None]
|
|
if self.use_scale_shift_norm:
|
|
out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
|
|
scale, shift = th.chunk(emb_out, 2, dim=1)
|
|
h = out_norm(h) * (1 + scale) + shift
|
|
h = out_rest(h)
|
|
else:
|
|
if self.exchange_temb_dims:
|
|
emb_out = rearrange(emb_out, "b t c ... -> b c t ...")
|
|
h = h + emb_out
|
|
h = self.out_layers(h)
|
|
return self.skip_connection(x) + h
|
|
|
|
|
|
class AttentionBlock(nn.Module):
|
|
"""
|
|
An attention block that allows spatial positions to attend to each other.
|
|
Originally ported from here, but adapted to the N-d case.
|
|
https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
channels: int,
|
|
num_heads: int = 1,
|
|
num_head_channels: int = -1,
|
|
use_checkpoint: bool = False,
|
|
use_new_attention_order: bool = False,
|
|
):
|
|
super().__init__()
|
|
self.channels = channels
|
|
if num_head_channels == -1:
|
|
self.num_heads = num_heads
|
|
else:
|
|
assert (
|
|
channels % num_head_channels == 0
|
|
), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
|
|
self.num_heads = channels // num_head_channels
|
|
self.use_checkpoint = use_checkpoint
|
|
self.norm = normalization(channels)
|
|
self.qkv = conv_nd(1, channels, channels * 3, 1)
|
|
if use_new_attention_order:
|
|
# split qkv before split heads
|
|
self.attention = QKVAttention(self.num_heads)
|
|
else:
|
|
# split heads before split qkv
|
|
self.attention = QKVAttentionLegacy(self.num_heads)
|
|
|
|
self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
|
|
|
|
def forward(self, x: th.Tensor, **kwargs) -> th.Tensor:
|
|
return checkpoint(self._forward, x)
|
|
|
|
def _forward(self, x: th.Tensor) -> th.Tensor:
|
|
b, c, *spatial = x.shape
|
|
x = x.reshape(b, c, -1)
|
|
qkv = self.qkv(self.norm(x))
|
|
h = self.attention(qkv)
|
|
h = self.proj_out(h)
|
|
return (x + h).reshape(b, c, *spatial)
|
|
|
|
|
|
class QKVAttentionLegacy(nn.Module):
|
|
"""
|
|
A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
|
|
"""
|
|
|
|
def __init__(self, n_heads: int):
|
|
super().__init__()
|
|
self.n_heads = n_heads
|
|
|
|
def forward(self, qkv: th.Tensor) -> th.Tensor:
|
|
"""
|
|
Apply QKV attention.
|
|
:param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
|
|
:return: an [N x (H * C) x T] tensor after attention.
|
|
"""
|
|
bs, width, length = qkv.shape
|
|
assert width % (3 * self.n_heads) == 0
|
|
ch = width // (3 * self.n_heads)
|
|
q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
|
|
scale = 1 / math.sqrt(math.sqrt(ch))
|
|
weight = th.einsum(
|
|
"bct,bcs->bts", q * scale, k * scale
|
|
) # More stable with f16 than dividing afterwards
|
|
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
|
a = th.einsum("bts,bcs->bct", weight, v)
|
|
return a.reshape(bs, -1, length)
|
|
|
|
|
|
class QKVAttention(nn.Module):
|
|
"""
|
|
A module which performs QKV attention and splits in a different order.
|
|
"""
|
|
|
|
def __init__(self, n_heads: int):
|
|
super().__init__()
|
|
self.n_heads = n_heads
|
|
|
|
def forward(self, qkv: th.Tensor) -> th.Tensor:
|
|
"""
|
|
Apply QKV attention.
|
|
:param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs.
|
|
:return: an [N x (H * C) x T] tensor after attention.
|
|
"""
|
|
bs, width, length = qkv.shape
|
|
assert width % (3 * self.n_heads) == 0
|
|
ch = width // (3 * self.n_heads)
|
|
q, k, v = qkv.chunk(3, dim=1)
|
|
scale = 1 / math.sqrt(math.sqrt(ch))
|
|
weight = th.einsum(
|
|
"bct,bcs->bts",
|
|
(q * scale).view(bs * self.n_heads, ch, length),
|
|
(k * scale).view(bs * self.n_heads, ch, length),
|
|
) # More stable with f16 than dividing afterwards
|
|
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
|
a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length))
|
|
return a.reshape(bs, -1, length)
|
|
|
|
|
|
class Timestep(nn.Module):
|
|
def __init__(self, dim: int):
|
|
super().__init__()
|
|
self.dim = dim
|
|
|
|
def forward(self, t: th.Tensor) -> th.Tensor:
|
|
return timestep_embedding(t, self.dim)
|
|
|
|
|
|
class UNetModel(nn.Module):
|
|
"""
|
|
The full UNet model with attention and timestep embedding.
|
|
:param in_channels: channels in the input Tensor.
|
|
:param model_channels: base channel count for the model.
|
|
:param out_channels: channels in the output Tensor.
|
|
:param num_res_blocks: number of residual blocks per downsample.
|
|
:param attention_resolutions: a collection of downsample rates at which
|
|
attention will take place. May be a set, list, or tuple.
|
|
For example, if this contains 4, then at 4x downsampling, attention
|
|
will be used.
|
|
:param dropout: the dropout probability.
|
|
:param channel_mult: channel multiplier for each level of the UNet.
|
|
:param conv_resample: if True, use learned convolutions for upsampling and
|
|
downsampling.
|
|
:param dims: determines if the signal is 1D, 2D, or 3D.
|
|
:param num_classes: if specified (as an int), then this model will be
|
|
class-conditional with `num_classes` classes.
|
|
:param use_checkpoint: use gradient checkpointing to reduce memory usage.
|
|
:param num_heads: the number of attention heads in each attention layer.
|
|
:param num_heads_channels: if specified, ignore num_heads and instead use
|
|
a fixed channel width per attention head.
|
|
:param num_heads_upsample: works with num_heads to set a different number
|
|
of heads for upsampling. Deprecated.
|
|
:param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
|
|
:param resblock_updown: use residual blocks for up/downsampling.
|
|
:param use_new_attention_order: use a different attention pattern for potentially
|
|
increased efficiency.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
in_channels: int,
|
|
model_channels: int,
|
|
out_channels: int,
|
|
num_res_blocks: int,
|
|
attention_resolutions: int,
|
|
dropout: float = 0.0,
|
|
channel_mult: Union[List, Tuple] = (1, 2, 4, 8),
|
|
conv_resample: bool = True,
|
|
dims: int = 2,
|
|
num_classes: Optional[Union[int, str]] = None,
|
|
use_checkpoint: bool = False,
|
|
num_heads: int = -1,
|
|
num_head_channels: int = -1,
|
|
num_heads_upsample: int = -1,
|
|
use_scale_shift_norm: bool = False,
|
|
resblock_updown: bool = False,
|
|
transformer_depth: int = 1,
|
|
context_dim: Optional[int] = None,
|
|
disable_self_attentions: Optional[List[bool]] = None,
|
|
num_attention_blocks: Optional[List[int]] = None,
|
|
disable_middle_self_attn: bool = False,
|
|
disable_middle_transformer: bool = False,
|
|
use_linear_in_transformer: bool = False,
|
|
spatial_transformer_attn_type: str = "softmax",
|
|
adm_in_channels: Optional[int] = None,
|
|
):
|
|
super().__init__()
|
|
|
|
if num_heads_upsample == -1:
|
|
num_heads_upsample = num_heads
|
|
|
|
if num_heads == -1:
|
|
assert (
|
|
num_head_channels != -1
|
|
), "Either num_heads or num_head_channels has to be set"
|
|
|
|
if num_head_channels == -1:
|
|
assert (
|
|
num_heads != -1
|
|
), "Either num_heads or num_head_channels has to be set"
|
|
|
|
self.in_channels = in_channels
|
|
self.model_channels = model_channels
|
|
self.out_channels = out_channels
|
|
if isinstance(transformer_depth, int):
|
|
transformer_depth = len(channel_mult) * [transformer_depth]
|
|
transformer_depth_middle = transformer_depth[-1]
|
|
|
|
if isinstance(num_res_blocks, int):
|
|
self.num_res_blocks = len(channel_mult) * [num_res_blocks]
|
|
else:
|
|
if len(num_res_blocks) != len(channel_mult):
|
|
msg = "provide num_res_blocks either as an int (globally constant) or as a list/tuple (per-level) with the same length as channel_mult"
|
|
raise ValueError(msg)
|
|
self.num_res_blocks = num_res_blocks
|
|
|
|
if disable_self_attentions is not None:
|
|
assert len(disable_self_attentions) == len(channel_mult)
|
|
if num_attention_blocks is not None:
|
|
assert len(num_attention_blocks) == len(self.num_res_blocks)
|
|
assert all(
|
|
self.num_res_blocks[i] >= num_attention_blocks[i]
|
|
for i in range(len(num_attention_blocks))
|
|
)
|
|
# logger.info(
|
|
# f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
|
|
# f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
|
|
# f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
|
|
# f"attention will still not be set."
|
|
# )
|
|
|
|
self.attention_resolutions = attention_resolutions
|
|
self.dropout = dropout
|
|
self.channel_mult = channel_mult
|
|
self.conv_resample = conv_resample
|
|
self.num_classes = num_classes
|
|
self.use_checkpoint = use_checkpoint
|
|
self.num_heads = num_heads
|
|
self.num_head_channels = num_head_channels
|
|
self.num_heads_upsample = num_heads_upsample
|
|
|
|
time_embed_dim = model_channels * 4
|
|
self.time_embed = nn.Sequential(
|
|
linear(model_channels, time_embed_dim),
|
|
nn.SiLU(),
|
|
linear(time_embed_dim, time_embed_dim),
|
|
)
|
|
|
|
if self.num_classes is not None:
|
|
if isinstance(self.num_classes, int):
|
|
self.label_emb = nn.Embedding(num_classes, time_embed_dim)
|
|
elif self.num_classes == "continuous":
|
|
# logger.debug("setting up linear c_adm embedding layer")
|
|
self.label_emb = nn.Linear(1, time_embed_dim)
|
|
elif self.num_classes == "timestep":
|
|
self.label_emb = nn.Sequential(
|
|
Timestep(model_channels),
|
|
nn.Sequential(
|
|
linear(model_channels, time_embed_dim),
|
|
nn.SiLU(),
|
|
linear(time_embed_dim, time_embed_dim),
|
|
),
|
|
)
|
|
elif self.num_classes == "sequential":
|
|
assert adm_in_channels is not None
|
|
self.label_emb = nn.Sequential(
|
|
nn.Sequential(
|
|
linear(adm_in_channels, time_embed_dim),
|
|
nn.SiLU(),
|
|
linear(time_embed_dim, time_embed_dim),
|
|
)
|
|
)
|
|
else:
|
|
raise ValueError
|
|
|
|
self.input_blocks = nn.ModuleList(
|
|
[
|
|
TimestepEmbedSequential(
|
|
conv_nd(dims, in_channels, model_channels, 3, padding=1)
|
|
)
|
|
]
|
|
)
|
|
self._feature_size = model_channels
|
|
input_block_chans = [model_channels]
|
|
ch = model_channels
|
|
ds = 1
|
|
for level, mult in enumerate(channel_mult):
|
|
for nr in range(self.num_res_blocks[level]):
|
|
layers = [
|
|
ResBlock(
|
|
ch,
|
|
time_embed_dim,
|
|
dropout,
|
|
out_channels=mult * model_channels,
|
|
dims=dims,
|
|
use_checkpoint=use_checkpoint,
|
|
use_scale_shift_norm=use_scale_shift_norm,
|
|
)
|
|
]
|
|
ch = mult * model_channels
|
|
if ds in attention_resolutions:
|
|
if num_head_channels == -1:
|
|
dim_head = ch // num_heads
|
|
else:
|
|
num_heads = ch // num_head_channels
|
|
dim_head = num_head_channels
|
|
|
|
if context_dim is not None and exists(disable_self_attentions):
|
|
disabled_sa = disable_self_attentions[level]
|
|
else:
|
|
disabled_sa = False
|
|
|
|
if (
|
|
not exists(num_attention_blocks)
|
|
or nr < num_attention_blocks[level]
|
|
):
|
|
layers.append(
|
|
SpatialTransformer(
|
|
ch,
|
|
num_heads,
|
|
dim_head,
|
|
depth=transformer_depth[level],
|
|
context_dim=context_dim,
|
|
disable_self_attn=disabled_sa,
|
|
use_linear=use_linear_in_transformer,
|
|
attn_type=spatial_transformer_attn_type,
|
|
use_checkpoint=use_checkpoint,
|
|
)
|
|
)
|
|
self.input_blocks.append(TimestepEmbedSequential(*layers))
|
|
self._feature_size += ch
|
|
input_block_chans.append(ch)
|
|
if level != len(channel_mult) - 1:
|
|
out_ch = ch
|
|
self.input_blocks.append(
|
|
TimestepEmbedSequential(
|
|
ResBlock(
|
|
ch,
|
|
time_embed_dim,
|
|
dropout,
|
|
out_channels=out_ch,
|
|
dims=dims,
|
|
use_checkpoint=use_checkpoint,
|
|
use_scale_shift_norm=use_scale_shift_norm,
|
|
down=True,
|
|
)
|
|
if resblock_updown
|
|
else Downsample(
|
|
ch, conv_resample, dims=dims, out_channels=out_ch
|
|
)
|
|
)
|
|
)
|
|
ch = out_ch
|
|
input_block_chans.append(ch)
|
|
ds *= 2
|
|
self._feature_size += ch
|
|
|
|
if num_head_channels == -1:
|
|
dim_head = ch // num_heads
|
|
else:
|
|
num_heads = ch // num_head_channels
|
|
dim_head = num_head_channels
|
|
|
|
self.middle_block = TimestepEmbedSequential(
|
|
ResBlock(
|
|
ch,
|
|
time_embed_dim,
|
|
dropout,
|
|
out_channels=ch,
|
|
dims=dims,
|
|
use_checkpoint=use_checkpoint,
|
|
use_scale_shift_norm=use_scale_shift_norm,
|
|
),
|
|
SpatialTransformer(
|
|
ch,
|
|
num_heads,
|
|
dim_head,
|
|
depth=transformer_depth_middle,
|
|
context_dim=context_dim,
|
|
disable_self_attn=disable_middle_self_attn,
|
|
use_linear=use_linear_in_transformer,
|
|
attn_type=spatial_transformer_attn_type,
|
|
use_checkpoint=use_checkpoint,
|
|
)
|
|
if not disable_middle_transformer
|
|
else th.nn.Identity(),
|
|
ResBlock(
|
|
ch,
|
|
time_embed_dim,
|
|
dropout,
|
|
dims=dims,
|
|
use_checkpoint=use_checkpoint,
|
|
use_scale_shift_norm=use_scale_shift_norm,
|
|
),
|
|
)
|
|
self._feature_size += ch
|
|
|
|
self.output_blocks = nn.ModuleList([])
|
|
for level, mult in list(enumerate(channel_mult))[::-1]:
|
|
for i in range(self.num_res_blocks[level] + 1):
|
|
ich = input_block_chans.pop()
|
|
layers = [
|
|
ResBlock(
|
|
ch + ich,
|
|
time_embed_dim,
|
|
dropout,
|
|
out_channels=model_channels * mult,
|
|
dims=dims,
|
|
use_checkpoint=use_checkpoint,
|
|
use_scale_shift_norm=use_scale_shift_norm,
|
|
)
|
|
]
|
|
ch = model_channels * mult
|
|
if ds in attention_resolutions:
|
|
if num_head_channels == -1:
|
|
dim_head = ch // num_heads
|
|
else:
|
|
num_heads = ch // num_head_channels
|
|
dim_head = num_head_channels
|
|
|
|
if exists(disable_self_attentions):
|
|
disabled_sa = disable_self_attentions[level]
|
|
else:
|
|
disabled_sa = False
|
|
|
|
if (
|
|
not exists(num_attention_blocks)
|
|
or i < num_attention_blocks[level]
|
|
):
|
|
layers.append(
|
|
SpatialTransformer(
|
|
ch,
|
|
num_heads,
|
|
dim_head,
|
|
depth=transformer_depth[level],
|
|
context_dim=context_dim,
|
|
disable_self_attn=disabled_sa,
|
|
use_linear=use_linear_in_transformer,
|
|
attn_type=spatial_transformer_attn_type,
|
|
use_checkpoint=use_checkpoint,
|
|
)
|
|
)
|
|
if level and i == self.num_res_blocks[level]:
|
|
out_ch = ch
|
|
layers.append(
|
|
ResBlock(
|
|
ch,
|
|
time_embed_dim,
|
|
dropout,
|
|
out_channels=out_ch,
|
|
dims=dims,
|
|
use_checkpoint=use_checkpoint,
|
|
use_scale_shift_norm=use_scale_shift_norm,
|
|
up=True,
|
|
)
|
|
if resblock_updown
|
|
else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
|
|
)
|
|
ds //= 2
|
|
self.output_blocks.append(TimestepEmbedSequential(*layers))
|
|
self._feature_size += ch
|
|
|
|
self.out = nn.Sequential(
|
|
normalization(ch),
|
|
nn.SiLU(),
|
|
zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)),
|
|
)
|
|
|
|
def forward(
|
|
self,
|
|
x: th.Tensor,
|
|
timesteps: Optional[th.Tensor] = None,
|
|
context: Optional[th.Tensor] = None,
|
|
y: Optional[th.Tensor] = None,
|
|
**kwargs,
|
|
) -> th.Tensor:
|
|
"""
|
|
Apply the model to an input batch.
|
|
:param x: an [N x C x ...] Tensor of inputs.
|
|
:param timesteps: a 1-D batch of timesteps.
|
|
:param context: conditioning plugged in via crossattn
|
|
:param y: an [N] Tensor of labels, if class-conditional.
|
|
:return: an [N x C x ...] Tensor of outputs.
|
|
"""
|
|
assert (y is not None) == (
|
|
self.num_classes is not None
|
|
), "must specify y if and only if the model is class-conditional"
|
|
hs = []
|
|
t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
|
|
emb = self.time_embed(t_emb)
|
|
|
|
if self.num_classes is not None:
|
|
assert y.shape[0] == x.shape[0]
|
|
emb = emb + self.label_emb(y)
|
|
|
|
h = x
|
|
for module in self.input_blocks:
|
|
h = module(h, emb, context)
|
|
hs.append(h)
|
|
h = self.middle_block(h, emb, context)
|
|
for module in self.output_blocks:
|
|
h = th.cat([h, hs.pop()], dim=1)
|
|
h = module(h, emb, context)
|
|
h = h.type(x.dtype)
|
|
|
|
return self.out(h)
|