Spaces:
Runtime error
Runtime error
File size: 12,730 Bytes
9667e74 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 |
import sys
import math
import torch
from torch import nn
from torch.nn import functional as F
from NTED.op import FusedLeakyReLU, fused_leaky_relu, upfirdn2d, conv2d_gradfix
class ExtractionOperation(nn.Module):
def __init__(self, in_channel, num_label, match_kernel):
super(ExtractionOperation, self).__init__()
self.value_conv = EqualConv2d(in_channel, in_channel, match_kernel, 1, match_kernel//2, bias=True)
self.semantic_extraction_filter = EqualConv2d(in_channel, num_label, match_kernel, 1, match_kernel//2, bias=False)
self.softmax = nn.Softmax(dim=-1)
self.num_label = num_label
def forward(self, value, recoder):
key = value
b,c,h,w = value.shape
key = self.semantic_extraction_filter(self.feature_norm(key))
extraction_softmax = self.softmax(key.view(b, -1, h*w)) #bkm
values_flatten = self.value_conv(value).view(b, -1, h*w)
neural_textures = torch.einsum('bkm,bvm->bvk', extraction_softmax, values_flatten)
recoder['extraction_softmax'].insert(0, extraction_softmax)
recoder['neural_textures'].insert(0, neural_textures)
return neural_textures, extraction_softmax
def feature_norm(self, input_tensor):
input_tensor = input_tensor - input_tensor.mean(dim=1, keepdim=True)
norm = torch.norm(input_tensor, 2, 1, keepdim=True) + sys.float_info.epsilon
out = torch.div(input_tensor, norm)
return out
class DistributionOperation(nn.Module):
def __init__(self, num_label, input_dim, match_kernel=3):
super(DistributionOperation, self).__init__()
self.semantic_distribution_filter = EqualConv2d(input_dim, num_label,
kernel_size=match_kernel,
stride=1,
padding=match_kernel//2)
self.num_label = num_label
def forward(self, query, extracted_feature, recoder):
b,c,h,w = query.shape
query = self.semantic_distribution_filter(query)
query_flatten = query.view(b, self.num_label, -1)
query_softmax = F.softmax(query_flatten, 1)
values_q = torch.einsum('bkm,bkv->bvm', query_softmax, extracted_feature.permute(0,2,1))
attn_out = values_q.view(b,-1,h,w)
recoder['semantic_distribution'].append(query)
return attn_out
class EncoderLayer(nn.Sequential):
def __init__(
self,
in_channel,
out_channel,
kernel_size,
downsample=False,
blur_kernel=[1, 3, 3, 1],
bias=True,
activate=True,
use_extraction=False,
num_label=None,
match_kernel=None,
num_extractions=2
):
super().__init__()
if downsample:
factor = 2
p = (len(blur_kernel) - factor) + (kernel_size - 1)
pad0 = (p + 1) // 2
pad1 = p // 2
self.blur = Blur(blur_kernel, pad=(pad0, pad1))
stride = 2
padding = 0
else:
self.blur = None
stride = 1
padding = kernel_size // 2
self.conv = EqualConv2d(
in_channel,
out_channel,
kernel_size,
padding=padding,
stride=stride,
bias=bias and not activate,
)
self.activate = FusedLeakyReLU(out_channel, bias=bias) if activate else None
self.use_extraction = use_extraction
if self.use_extraction:
self.extraction_operations = nn.ModuleList()
for _ in range(num_extractions):
self.extraction_operations.append(
ExtractionOperation(
out_channel,
num_label,
match_kernel
)
)
def forward(self, input, recoder=None):
out = self.blur(input) if self.blur is not None else input
out = self.conv(out)
out = self.activate(out) if self.activate is not None else out
if self.use_extraction:
for extraction_operation in self.extraction_operations:
extraction_operation(out, recoder)
return out
class DecoderLayer(nn.Module):
def __init__(
self,
in_channel,
out_channel,
kernel_size,
upsample=False,
blur_kernel=[1, 3, 3, 1],
bias=True,
activate=True,
use_distribution=True,
num_label=16,
match_kernel=3,
):
super().__init__()
if upsample:
factor = 2
p = (len(blur_kernel) - factor) - (kernel_size - 1)
pad0 = (p + 1) // 2 + factor - 1
pad1 = p // 2 + 1
self.blur = Blur(blur_kernel, pad=(pad0, pad1), upsample_factor=factor)
self.conv = EqualTransposeConv2d(
in_channel,
out_channel,
kernel_size,
stride=2,
padding=0,
bias=bias and not activate,
)
else:
self.conv = EqualConv2d(
in_channel,
out_channel,
kernel_size,
stride=1,
padding=kernel_size//2,
bias=bias and not activate,
)
self.blur = None
self.distribution_operation = DistributionOperation(
num_label,
out_channel,
match_kernel=match_kernel
) if use_distribution else None
self.activate = FusedLeakyReLU(out_channel, bias=bias) if activate else None
self.use_distribution = use_distribution
def forward(self, input, neural_texture=None, recoder=None):
out = self.conv(input)
out = self.blur(out) if self.blur is not None else out
if self.use_distribution and neural_texture is not None:
out_attn = self.distribution_operation(out, neural_texture, recoder)
out = (out + out_attn) / math.sqrt(2)
out = self.activate(out.contiguous()) if self.activate is not None else out
return out
class EqualConv2d(nn.Module):
def __init__(
self, in_channel, out_channel, kernel_size, stride=1, padding=0, bias=True
):
super().__init__()
self.weight = nn.Parameter(
torch.randn(out_channel, in_channel, kernel_size, kernel_size)
)
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
def forward(self, input):
out = conv2d_gradfix.conv2d(
input,
self.weight * self.scale,
bias=self.bias,
stride=self.stride,
padding=self.padding,
)
return out
def __repr__(self):
return (
f"{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]},"
f" {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})"
)
class EqualTransposeConv2d(nn.Module):
def __init__(
self, in_channel, out_channel, kernel_size, stride=1, padding=0, bias=True
):
super().__init__()
self.weight = nn.Parameter(
torch.randn(out_channel, in_channel, kernel_size, kernel_size)
)
self.scale = 1 / math.sqrt(in_channel * kernel_size ** 2)
self.stride = stride
self.padding = padding
if bias:
self.bias = nn.Parameter(torch.zeros(out_channel))
else:
self.bias = None
def forward(self, input):
weight = self.weight.transpose(0,1)
out = conv2d_gradfix.conv_transpose2d(
input,
weight * self.scale,
bias=self.bias,
stride=self.stride,
padding=self.padding,
)
return out
def __repr__(self):
return (
f"{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]},"
f" {self.weight.shape[2]}, stride={self.stride}, padding={self.padding})"
)
class ToRGB(nn.Module):
def __init__(
self,
in_channel,
upsample=True,
blur_kernel=[1, 3, 3, 1]
):
super().__init__()
if upsample:
self.upsample = Upsample(blur_kernel)
self.conv = EqualConv2d(in_channel, 3, 3, stride=1, padding=1)
def forward(self, input, skip=None):
out = self.conv(input)
if skip is not None:
skip = self.upsample(skip)
out = out + skip
return out
class EqualLinear(nn.Module):
def __init__(
self, in_dim, out_dim, bias=True, bias_init=0, lr_mul=1, activation=None
):
super().__init__()
self.weight = nn.Parameter(torch.randn(out_dim, in_dim).div_(lr_mul))
if bias:
self.bias = nn.Parameter(torch.zeros(out_dim).fill_(bias_init))
else:
self.bias = None
self.activation = activation
self.scale = (1 / math.sqrt(in_dim)) * lr_mul
self.lr_mul = lr_mul
def forward(self, input):
if self.activation:
out = F.linear(input, self.weight * self.scale)
out = fused_leaky_relu(out, self.bias * self.lr_mul)
else:
out = F.linear(
input, self.weight * self.scale, bias=self.bias * self.lr_mul
)
return out
def __repr__(self):
return (
f"{self.__class__.__name__}({self.weight.shape[1]}, {self.weight.shape[0]})"
)
class Upsample(nn.Module):
def __init__(self, kernel, factor=2):
super().__init__()
self.factor = factor
kernel = make_kernel(kernel) * (factor ** 2)
self.register_buffer("kernel", kernel)
p = kernel.shape[0] - factor
pad0 = (p + 1) // 2 + factor - 1
pad1 = p // 2
self.pad = (pad0, pad1)
def forward(self, input):
out = upfirdn2d(input, self.kernel, up=self.factor, down=1, pad=self.pad)
return out
class ResBlock(nn.Module):
def __init__(self, in_channel, out_channel, blur_kernel=[1, 3, 3, 1]):
super().__init__()
self.conv1 = ConvLayer(in_channel, in_channel, 3)
self.conv2 = ConvLayer(in_channel, out_channel, 3, downsample=True)
self.skip = ConvLayer(
in_channel, out_channel, 1, downsample=True, activate=False, bias=False
)
def forward(self, input):
out = self.conv1(input)
out = self.conv2(out)
skip = self.skip(input)
out = (out + skip) / math.sqrt(2)
return out
class ConvLayer(nn.Sequential):
def __init__(
self,
in_channel,
out_channel,
kernel_size,
downsample=False,
blur_kernel=[1, 3, 3, 1],
bias=True,
activate=True,
):
layers = []
if downsample:
factor = 2
p = (len(blur_kernel) - factor) + (kernel_size - 1)
pad0 = (p + 1) // 2
pad1 = p // 2
layers.append(Blur(blur_kernel, pad=(pad0, pad1)))
stride = 2
self.padding = 0
else:
stride = 1
self.padding = kernel_size // 2
layers.append(
EqualConv2d(
in_channel,
out_channel,
kernel_size,
padding=self.padding,
stride=stride,
bias=bias and not activate,
)
)
if activate:
layers.append(FusedLeakyReLU(out_channel, bias=bias))
super().__init__(*layers)
class Blur(nn.Module):
def __init__(self, kernel, pad, upsample_factor=1):
super().__init__()
kernel = make_kernel(kernel)
if upsample_factor > 1:
kernel = kernel * (upsample_factor ** 2)
self.register_buffer("kernel", kernel)
self.pad = pad
def forward(self, input):
out = upfirdn2d(input, self.kernel, pad=self.pad)
return out
def make_kernel(k):
k = torch.tensor(k, dtype=torch.float32)
if k.ndim == 1:
k = k[None, :] * k[:, None]
k /= k.sum()
return k
def accumulate(model1, model2, decay=0.999):
par1 = dict(model1.named_parameters())
par2 = dict(model2.named_parameters())
for k in par1.keys():
par1[k].data.mul_(decay).add_(par2[k].data, alpha=1 - decay) |