File size: 8,138 Bytes
998bb30 |
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 |
import math
from telnetlib import PRAGMA_HEARTBEAT
import torch
from torch import nn
from torch.nn import init
from torch.nn import functional as F
from einops import rearrange
class TargetEmbedding(nn.Module):
def __init__(self, dim, feature_channel_num):
super().__init__()
self.linear = nn.Sequential(
nn.Linear(feature_channel_num, dim),
nn.GELU(),
nn.Linear(dim, dim),
)
def forward(self,mix):
mix_linear = self.linear(mix)
return mix_linear
class DownSample(nn.Module):
def __init__(self, in_ch):
super().__init__()
self.c1 = nn.Sequential(nn.GELU(),
nn.Conv2d(in_ch, in_ch, 3, stride=2, padding=1, padding_mode='reflect'))
self.c2 = nn.Sequential(nn.GELU(),
nn.Conv2d(in_ch, in_ch, 5, stride=2, padding=2, padding_mode='reflect'))
def forward(self, x, target_emb):
x = self.c1(x)+self.c2(x)
return x
class UpSample(nn.Module):
def __init__(self, in_ch):
super().__init__()
self.t = nn.Sequential(nn.GELU(), nn.GroupNorm(16, in_ch),
nn.ConvTranspose2d(in_ch, in_ch, 5, 2, 2, 1))
def forward(self, x, target_emb):
_, _, H, W = x.shape
x = self.t(x)
return x
class AttnBlock(nn.Module):
def __init__(self, in_ch):
super().__init__()
self.norm = nn.GroupNorm(16,in_ch)
self.to_kv=nn.Conv2d(in_ch,in_ch*2,1)
self.out = nn.Sequential(nn.GroupNorm(16,in_ch),
nn.GELU(),
nn.Conv2d(in_ch, in_ch, 1, stride=1, padding=0))
def forward(self, x):
B, C, H, W = x.shape
x = self.norm(x)
q_scale = int(C) ** (-0.5)
kv = self.to_kv(x).chunk(2,dim=1)
k,v = map(lambda t: rearrange(t, 'b c x y -> b c (x y)'),kv)
q = F.softmax(k,dim=-2)
k = F.softmax(k,dim=-1)
q = q*q_scale
context = torch.einsum('b d n, b e n -> b d e',k,v)
assert list(context.shape) == [B, C, C]
out = torch.einsum('b d e, b d n -> b e n',context,q)
assert list(out.shape) == [B, C, H*W]
out = rearrange(out,'b c (i j) -> b c i j', i=H ,j=W)
out = self.out(out)
return x + out
class ResBlock(nn.Module):
def __init__(self, in_ch, out_ch, tdim, attn=True):
super().__init__()
self.block1 = nn.Sequential(
nn.GELU(),
nn.GroupNorm(16, in_ch),
nn.Conv2d(in_ch, out_ch, 3, stride=1, padding=1, padding_mode='reflect'),
)
self.target_proj = nn.Sequential(
nn.GELU(),
nn.Linear(tdim, out_ch),
)
self.block2 = nn.Sequential(
nn.GELU(),
nn.GroupNorm(16, out_ch),
nn.Conv2d(out_ch, out_ch, 3, stride=1, padding=1, padding_mode='reflect'),
)
if in_ch != out_ch:
self.shortcut = nn.Conv2d(in_ch, out_ch, 1, stride=1, padding=0)
else:
self.shortcut = nn.Identity()
if attn:
self.attn = AttnBlock(out_ch)
else:
self.attn = nn.Identity()
def forward(self, x , target):
h = self.block1(x)
h += self.target_proj(target)[:, :, None, None]
h = self.block2(h)
B,C,H,W = h.size()
layer_norm = nn.LayerNorm([C,H,W])
h = layer_norm.to(h.device)(h)
h = h + self.shortcut(x)
h = self.attn(h)
return h
class GCT(nn.Module):
def __init__(self, num_channels, tdim, epsilon=1e-5, mode='l2', after_relu=False):
super(GCT, self).__init__()
self.alpha = nn.Parameter(torch.ones(1, num_channels, 1, 1))
self.gamma = nn.Parameter(torch.zeros(1, num_channels, 1, 1))
self.beta = nn.Parameter(torch.zeros(1, num_channels, 1, 1))
self.epsilon = epsilon
self.mode = mode
self.after_relu = after_relu
def forward(self, x):
b, c, h, w = x.shape
if self.mode == 'l2':
embedding = (x.pow(2).sum((2, 3), keepdim=True) +
self.epsilon).pow(0.5) * self.alpha
norm = self.gamma / \
(embedding.pow(2).mean(dim=1, keepdim=True) + self.epsilon).pow(0.5)
elif self.mode == 'l1':
if not self.after_relu:
_x = torch.abs(x)
else:
_x = x
embedding = _x.sum((2, 3), keepdim=True) * self.alpha
norm = self.gamma / \
(torch.abs(embedding).mean(dim=1, keepdim=True) + self.epsilon)
gate = 1. + torch.tanh(embedding * norm + self.beta)
return x * gate
class Generator(nn.Module):
def __init__(self, num_target,feature_channel_num, ch, ch_mult, num_res_blocks,inception=False):
super().__init__()
tdim = ch * 4
self.target_embedding = TargetEmbedding(tdim,feature_channel_num)
self.head = nn.Sequential(
nn.Conv2d(3, ch, kernel_size=5, padding=2, padding_mode='reflect'),
nn.GELU(),
nn.GroupNorm(16, ch)
)
self.crop = nn.ConstantPad2d((0, -1, -1, 0), 0)
self.inception = inception
self.downblocks = nn.ModuleList()
chs = [ch]
now_ch = ch
feature_channel = feature_channel_num
for i, mult in enumerate(ch_mult):
out_ch = ch * mult
for _ in range(num_res_blocks):
self.downblocks.append(ResBlock(in_ch=now_ch, out_ch=out_ch, tdim=tdim))
now_ch = out_ch
chs.append(now_ch)
if i != len(ch_mult) - 1:
self.downblocks.append(DownSample(now_ch))
chs.append(now_ch)
self.middleblocks = nn.ModuleList([
ResBlock(now_ch, now_ch, tdim, attn=True),
ResBlock(now_ch, now_ch, tdim, attn=False),
])
self.upblocks = nn.ModuleList()
for i, mult in reversed(list(enumerate(ch_mult))):
out_ch = ch * mult
for _ in range(num_res_blocks + 1):
self.upblocks.append(ResBlock(in_ch=chs.pop() + now_ch, out_ch=out_ch, tdim=tdim, attn=False))
now_ch = out_ch
if i != 0:
self.upblocks.append(UpSample(now_ch))
assert len(chs) == 0
self.gct = GCT(num_channels=now_ch,tdim=tdim)
self.tail = nn.Sequential(
nn.GroupNorm(16, now_ch),
nn.GELU(),
nn.Conv2d(now_ch, 3, 3, stride=1, padding=1, padding_mode='reflect'),
)
def weight_init(self):
for m in self.modules():
if isinstance(m,torch.nn.Conv2d):
init.xavier_uniform_(m.weight.data)
if m.bias is not None:
init.constant_(m.bias.data,0.1)
elif isinstance(m,torch.nn.GroupNorm):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m,torch.nn.Linear):
m.weight.data.normal_(0,0.01)
if m.bias is not None:
m.bias.data.zero_()
def forward(self, x,mix):
targetemb = self.target_embedding(mix)
h = self.head(x)
hs = [h]
for layer in self.downblocks:
h = layer(h, targetemb)
hs.append(h)
for layer in self.middleblocks:
h = layer(h, targetemb)
for layer in self.upblocks:
if isinstance(layer, ResBlock):
hs_pop = hs.pop()
if hs_pop.size(2) != h.size(2) and self.inception:
h = self.crop(h)
h = torch.cat([h, hs_pop], dim=1)
h = layer(h, targetemb)
h = self.gct(h)
h = self.tail(h)
assert len(hs) == 0
return (torch.tanh(h) + 1) / 2
|