-
Notifications
You must be signed in to change notification settings - Fork 18
/
model.py
140 lines (124 loc) · 5.08 KB
/
model.py
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
"""
Pytorch implementation of Self-Supervised GAN
Reference: "Self-Supervised GANs via Auxiliary Rotation Loss"
Authors: Ting Chen,
Xiaohua Zhai,
Marvin Ritter,
Mario Lucic and
Neil Houlsby
https://arxiv.org/abs/1811.11212 CVPR 2019.
Script Author: Vandit Jain. Github:vandit15
"""
import os.path
import tarfile, sys, math
from six.moves import urllib
from ops import conv2d, deconv2d, Residual_G, Residual_D
import numpy as np
import scipy
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
from torchvision.utils import make_grid
from torch.autograd import grad as torch_grad
class Discriminator(nn.Module):
def __init__(self, spectral_normed, num_rotation,
ssup, channel, resnet = False):
super(Discriminator, self).__init__()
self.resnet = resnet
self.num_rotation = num_rotation
self.ssup = ssup
self.lrelu = nn.LeakyReLU()
self.relu = nn.ReLU()
self.conv1 = conv2d(channel, 64, kernel_size = 3, stride = 1, padding = 1,
spectral_normed = spectral_normed)
self.conv2 = conv2d(64, 128, spectral_normed = spectral_normed,
padding = 0)
self.conv3 = conv2d(128, 256, spectral_normed = spectral_normed,
padding = 0)
self.conv4 = conv2d(256, 512, spectral_normed = spectral_normed,
padding = 0)
self.fully_connect_gan1 = nn.Linear(512, 1)
self.fully_connect_rot1 = nn.Linear(512, 4)
self.softmax = nn.Softmax()
self.re1 = Residual_D(channel, 128, spectral_normed = spectral_normed,
down_sampling = True, is_start = True)
self.re2 = Residual_D(128, 128, spectral_normed = spectral_normed,
down_sampling = True)
self.re3 = Residual_D(128, 128, spectral_normed = spectral_normed)
self.re4 = Residual_D(128, 128, spectral_normed = spectral_normed)
self.fully_connect_gan2 = nn.Linear(128, 1)
self.fully_connect_rot2 = nn.Linear(128, 4)
self.sigmoid = nn.Sigmoid()
def forward(self, x):
if self.resnet == False:
conv1 = self.lrelu(self.conv1(x))
conv2 = self.lrelu(self.conv2(conv1))
conv3 = self.lrelu(self.conv3(conv2))
conv4 = self.lrelu(self.conv4(conv3))
conv4 = torch.view(conv4.size(0)*self.num_rotation, -1)
gan_logits = self.fully_connect_gan1(conv4)
if self.ssup:
rot_logits = self.fully_connect_rot1(conv4)
rot_prob = self.softmax(rot_logits)
else:
re1 = self.re1(x)
re2 = self.re2(re1)
re3 = self.re3(re2)
re4 = self.re4(re3)
re4 = self.relu(re4)
re4 = torch.sum(re4,dim = (2,3))
gan_logits = self.fully_connect_gan2(re4)
if self.ssup:
rot_logits = self.fully_connect_rot2(re4)
rot_prob = self.softmax(rot_logits)
if self.ssup:
return self.sigmoid(gan_logits), gan_logits, rot_logits, rot_prob
else:
return self.sigmoid(gan_logits), gan_logits
class Generator(nn.Module):
def __init__(self, z_size, channel, resnet = False, output_size = 32):
super(Generator, self).__init__()
s = 4
self.output_size = output_size
if self.output_size == 32:
s = 4
if self.output_size == 48:
s = 6
self.s = s
self.z_size = z_size
self.resnet = resnet
self.fully_connect = nn.Linear(z_size, s*s*256)
self.relu = nn.ReLU()
self.tanh = nn.Tanh()
self.deconv1 = deconv2d(256, 256, padding = 0)
self.bn1 = nn.BatchNorm2d(256)
self.deconv2 = deconv2d(256, 128, padding = 0)
self.bn2 = nn.BatchNorm2d(128)
self.deconv3 = deconv2d(128, 64, padding = 0)
self.bn3 = nn.BatchNorm2d(64)
self.conv4 = conv2d(64, channel, padding = 1, kernel_size = 3, stride = 1)
self.conv_res4 = conv2d(256,channel, padding = 1, kernel_size = 3, stride = 1)
self.re1 = Residual_G(256, 256, up_sampling = True)
self.re2 = Residual_G(256, 256, up_sampling = True)
self.re3 = Residual_G(256, 256, up_sampling = True)
self.bn = nn.BatchNorm2d(256)
def forward(self, x):
d1 = self.fully_connect(x)
d1 = d1.view(-1, 256, self.s, self.s)
if self.resnet == False:
d1 = self.relu(d1)
d2 = self.relu(self.bn1(self.deconv1(d1)))
d3 = self.relu(self.bn2(self.deconv2(d2)))
d4 = self.relu(self.bn3(self.deconv3(d3)))
d5 = self.conv4(d4)
else:
d2 = self.re1(d1)
d3 = self.re2(d2)
d4 = self.re3(d3)
d4 = self.relu(self.bn(d4))
d5 = self.conv_res4(d4)
return self.tanh(d5)
def sample_latent(self, num_samples):
return torch.randn((num_samples, self.z_size))