File size: 13,559 Bytes
617d388 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 |
#taken from: https://github.com/lllyasviel/ControlNet
#and modified
import torch
import torch as th
import torch.nn as nn
from ldm_patched.ldm.modules.diffusionmodules.util import (
zero_module,
timestep_embedding,
)
from ldm_patched.ldm.modules.attention import SpatialTransformer
from ldm_patched.ldm.modules.diffusionmodules.openaimodel import UNetModel, TimestepEmbedSequential, ResBlock, Downsample
from ldm_patched.ldm.util import exists
import ldm_patched.modules.ops
class ControlledUnetModel(UNetModel):
#implemented in the ldm unet
pass
class ControlNet(nn.Module):
def __init__(
self,
image_size,
in_channels,
model_channels,
hint_channels,
num_res_blocks,
dropout=0,
channel_mult=(1, 2, 4, 8),
conv_resample=True,
dims=2,
num_classes=None,
use_checkpoint=False,
dtype=torch.float32,
num_heads=-1,
num_head_channels=-1,
num_heads_upsample=-1,
use_scale_shift_norm=False,
resblock_updown=False,
use_new_attention_order=False,
use_spatial_transformer=False, # custom transformer support
transformer_depth=1, # custom transformer support
context_dim=None, # custom transformer support
n_embed=None, # custom support for prediction of discrete ids into codebook of first stage vq model
legacy=True,
disable_self_attentions=None,
num_attention_blocks=None,
disable_middle_self_attn=False,
use_linear_in_transformer=False,
adm_in_channels=None,
transformer_depth_middle=None,
transformer_depth_output=None,
device=None,
operations=ldm_patched.modules.ops.disable_weight_init,
**kwargs,
):
super().__init__()
assert use_spatial_transformer == True, "use_spatial_transformer has to be true"
if use_spatial_transformer:
assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
if context_dim is not None:
assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
# from omegaconf.listconfig import ListConfig
# if type(context_dim) == ListConfig:
# context_dim = list(context_dim)
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.dims = dims
self.image_size = image_size
self.in_channels = in_channels
self.model_channels = model_channels
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):
raise ValueError("provide num_res_blocks either as an int (globally constant) or "
"as a list/tuple (per-level) with the same length as channel_mult")
self.num_res_blocks = num_res_blocks
if disable_self_attentions is not None:
# should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
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(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
transformer_depth = transformer_depth[:]
self.dropout = dropout
self.channel_mult = channel_mult
self.conv_resample = conv_resample
self.num_classes = num_classes
self.use_checkpoint = use_checkpoint
self.dtype = dtype
self.num_heads = num_heads
self.num_head_channels = num_head_channels
self.num_heads_upsample = num_heads_upsample
self.predict_codebook_ids = n_embed is not None
time_embed_dim = model_channels * 4
self.time_embed = nn.Sequential(
operations.Linear(model_channels, time_embed_dim, dtype=self.dtype, device=device),
nn.SiLU(),
operations.Linear(time_embed_dim, time_embed_dim, dtype=self.dtype, device=device),
)
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":
print("setting up linear c_adm embedding layer")
self.label_emb = nn.Linear(1, time_embed_dim)
elif self.num_classes == "sequential":
assert adm_in_channels is not None
self.label_emb = nn.Sequential(
nn.Sequential(
operations.Linear(adm_in_channels, time_embed_dim, dtype=self.dtype, device=device),
nn.SiLU(),
operations.Linear(time_embed_dim, time_embed_dim, dtype=self.dtype, device=device),
)
)
else:
raise ValueError()
self.input_blocks = nn.ModuleList(
[
TimestepEmbedSequential(
operations.conv_nd(dims, in_channels, model_channels, 3, padding=1, dtype=self.dtype, device=device)
)
]
)
self.zero_convs = nn.ModuleList([self.make_zero_conv(model_channels, operations=operations, dtype=self.dtype, device=device)])
self.input_hint_block = TimestepEmbedSequential(
operations.conv_nd(dims, hint_channels, 16, 3, padding=1, dtype=self.dtype, device=device),
nn.SiLU(),
operations.conv_nd(dims, 16, 16, 3, padding=1, dtype=self.dtype, device=device),
nn.SiLU(),
operations.conv_nd(dims, 16, 32, 3, padding=1, stride=2, dtype=self.dtype, device=device),
nn.SiLU(),
operations.conv_nd(dims, 32, 32, 3, padding=1, dtype=self.dtype, device=device),
nn.SiLU(),
operations.conv_nd(dims, 32, 96, 3, padding=1, stride=2, dtype=self.dtype, device=device),
nn.SiLU(),
operations.conv_nd(dims, 96, 96, 3, padding=1, dtype=self.dtype, device=device),
nn.SiLU(),
operations.conv_nd(dims, 96, 256, 3, padding=1, stride=2, dtype=self.dtype, device=device),
nn.SiLU(),
operations.conv_nd(dims, 256, model_channels, 3, padding=1, dtype=self.dtype, device=device)
)
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,
dtype=self.dtype,
device=device,
operations=operations,
)
]
ch = mult * model_channels
num_transformers = transformer_depth.pop(0)
if num_transformers > 0:
if num_head_channels == -1:
dim_head = ch // num_heads
else:
num_heads = ch // num_head_channels
dim_head = num_head_channels
if legacy:
#num_heads = 1
dim_head = ch // num_heads if use_spatial_transformer else 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 nr < num_attention_blocks[level]:
layers.append(
SpatialTransformer(
ch, num_heads, dim_head, depth=num_transformers, context_dim=context_dim,
disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
)
)
self.input_blocks.append(TimestepEmbedSequential(*layers))
self.zero_convs.append(self.make_zero_conv(ch, operations=operations, dtype=self.dtype, device=device))
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,
dtype=self.dtype,
device=device,
operations=operations
)
if resblock_updown
else Downsample(
ch, conv_resample, dims=dims, out_channels=out_ch, dtype=self.dtype, device=device, operations=operations
)
)
)
ch = out_ch
input_block_chans.append(ch)
self.zero_convs.append(self.make_zero_conv(ch, operations=operations, dtype=self.dtype, device=device))
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
if legacy:
#num_heads = 1
dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
mid_block = [
ResBlock(
ch,
time_embed_dim,
dropout,
dims=dims,
use_checkpoint=use_checkpoint,
use_scale_shift_norm=use_scale_shift_norm,
dtype=self.dtype,
device=device,
operations=operations
)]
if transformer_depth_middle >= 0:
mid_block += [SpatialTransformer( # always uses a self-attn
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,
use_checkpoint=use_checkpoint, dtype=self.dtype, device=device, operations=operations
),
ResBlock(
ch,
time_embed_dim,
dropout,
dims=dims,
use_checkpoint=use_checkpoint,
use_scale_shift_norm=use_scale_shift_norm,
dtype=self.dtype,
device=device,
operations=operations
)]
self.middle_block = TimestepEmbedSequential(*mid_block)
self.middle_block_out = self.make_zero_conv(ch, operations=operations, dtype=self.dtype, device=device)
self._feature_size += ch
def make_zero_conv(self, channels, operations=None, dtype=None, device=None):
return TimestepEmbedSequential(operations.conv_nd(self.dims, channels, channels, 1, padding=0, dtype=dtype, device=device))
def forward(self, x, hint, timesteps, context, y=None, **kwargs):
t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False).to(x.dtype)
emb = self.time_embed(t_emb)
guided_hint = self.input_hint_block(hint, emb, context)
outs = []
hs = []
if self.num_classes is not None:
assert y.shape[0] == x.shape[0]
emb = emb + self.label_emb(y)
h = x
for module, zero_conv in zip(self.input_blocks, self.zero_convs):
if guided_hint is not None:
h = module(h, emb, context)
h += guided_hint
guided_hint = None
else:
h = module(h, emb, context)
outs.append(zero_conv(h, emb, context))
h = self.middle_block(h, emb, context)
outs.append(self.middle_block_out(h, emb, context))
return outs
|