Waifu2x/waifu2x/srcnn.py at main · FHPythonUtils/Waifu2x · GitHub

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
from __future__ import annotations

import chainer.functions as F
import chainer.links as L
from chainer.link import Chain


class VGG7(Chain):
	def __init__(self, ch) -> None:
		super().__init__()
		with self.init_scope():
			self.conv1 = L.Convolution2D(ch, 32, 3)
			self.conv2 = L.Convolution2D(32, 32, 3)
			self.conv3 = L.Convolution2D(32, 64, 3)
			self.conv4 = L.Convolution2D(64, 64, 3)
			self.conv5 = L.Convolution2D(64, 128, 3)
			self.conv6 = L.Convolution2D(128, 128, 3)
			self.conv7 = L.Convolution2D(128, ch, 3)

		self.ch = ch
		self.offset = 7
		self.inner_scale = 1

	def __call__(self, x):
		h = F.leaky_relu(self.conv1(x), 0.1)
		h = F.leaky_relu(self.conv2(h), 0.1)
		h = F.leaky_relu(self.conv3(h), 0.1)
		h = F.leaky_relu(self.conv4(h), 0.1)
		h = F.leaky_relu(self.conv5(h), 0.1)
		h = F.leaky_relu(self.conv6(h), 0.1)
		return self.conv7(h)


class UpConv7(Chain):
	def __init__(self, ch: int) -> None:
		super().__init__()
		with self.init_scope():
			self.conv1 = L.Convolution2D(ch, 16, 3)
			self.conv2 = L.Convolution2D(16, 32, 3)
			self.conv3 = L.Convolution2D(32, 64, 3)
			self.conv4 = L.Convolution2D(64, 128, 3)
			self.conv5 = L.Convolution2D(128, 256, 3)
			self.conv6 = L.Convolution2D(256, 256, 3)
			self.conv7 = L.Deconvolution2D(256, ch, 4, 2, 3, nobias=True)

		self.ch = ch
		self.offset = 14
		self.inner_scale = 2

	def __call__(self, x):
		h = F.leaky_relu(self.conv1(x), 0.1)
		h = F.leaky_relu(self.conv2(h), 0.1)
		h = F.leaky_relu(self.conv3(h), 0.1)
		h = F.leaky_relu(self.conv4(h), 0.1)
		h = F.leaky_relu(self.conv5(h), 0.1)
		h = F.leaky_relu(self.conv6(h), 0.1)
		return self.conv7(h)


class ResBlock(Chain):
	def __init__(
		self, in_channels, out_channels, slope: float = 0.1, r: int = 16, se: bool = False
	) -> None:
		super().__init__()
		with self.init_scope():
			self.conv1 = L.Convolution2D(in_channels, out_channels, 3)
			self.conv2 = L.Convolution2D(out_channels, out_channels, 3)

			if in_channels != out_channels:
				self.conv_bridge = L.Convolution2D(in_channels, out_channels, 1)

			if se:
				self.fc1 = L.Linear(out_channels, out_channels // r)
				self.fc2 = L.Linear(out_channels // r, out_channels)

		self.slope = slope

	def __call__(self, x):
		h = F.leaky_relu(self.conv1(x), self.slope)
		h = F.leaky_relu(self.conv2(h), self.slope)

		if hasattr(self, "conv_bridge"):
			x = self.conv_bridge(x[:, :, 2:-2, 2:-2])
		else:
			x = x[:, :, 2:-2, 2:-2]

		if hasattr(self, "fc1") and hasattr(self, "fc2"):
			se = F.relu(self.fc1(F.average(h, axis=(2, 3))))
			se = F.sigmoid(self.fc2(se))[:, :, None, None]
			se = F.broadcast_to(se, h.shape)
			h = h * se

		return h + x


class ResNet10(Chain):
	def __init__(self, ch) -> None:
		super().__init__()
		with self.init_scope():
			self.conv_pre = L.Convolution2D(ch, 64, 3)
			self.res1 = ResBlock(64, 64)
			self.res2 = ResBlock(64, 64)
			self.res3 = ResBlock(64, 64)
			self.res4 = ResBlock(64, 64)
			self.res5 = ResBlock(64, 64)
			self.conv_bridge = L.Convolution2D(64, 64, 3)
			self.conv_post = L.Convolution2D(64, ch, 3)

		self.ch = ch
		self.offset = 13
		self.inner_scale = 1

	def __call__(self, x):
		h = bridge = F.leaky_relu(self.conv_pre(x), 0.1)
		h = self.res1(h)
		h = self.res2(h)
		h = self.res3(h)
		h = self.res4(h)
		h = self.res5(h)
		h = F.leaky_relu(self.conv_bridge(h), 0.1)
		h = h + bridge[:, :, 11:-11, 11:-11]
		return self.conv_post(h)


class UpResNet10(Chain):
	def __init__(self, ch) -> None:
		super().__init__()
		with self.init_scope():
			self.conv_pre = L.Convolution2D(ch, 64, 3)
			self.res1 = ResBlock(64, 64, r=4, se=True)
			self.res2 = ResBlock(64, 64, r=4, se=True)
			self.res3 = ResBlock(64, 64, r=4, se=True)
			self.res4 = ResBlock(64, 64, r=4, se=True)
			self.res5 = ResBlock(64, 64, r=4, se=True)
			self.conv_bridge = L.Convolution2D(64, 64, 3)
			self.conv_post = L.Deconvolution2D(64, ch, 4, 2, 3, nobias=True)

		self.ch = ch
		self.offset = 26
		self.inner_scale = 2

	def __call__(self, x):
		h = bridge = F.leaky_relu(self.conv_pre(x), 0.1)
		h = self.res1(h)
		h = self.res2(h)
		h = self.res3(h)
		h = self.res4(h)
		h = self.res5(h)
		h = F.leaky_relu(self.conv_bridge(h), 0.1)
		h = h + bridge[:, :, 11:-11, 11:-11]
		return self.conv_post(h)


archs = {
	"VGG7": VGG7,
	"UpConv7": UpConv7,
	"ResNet10": ResNet10,
	"UpResNet10": UpResNet10,
}

table = {
	"0": "VGG7",
	"1": "UpConv7",
	"2": "ResNet10",
	"3": "UpResNet10",
}