-
Notifications
You must be signed in to change notification settings - Fork 5
/
cppn.py
198 lines (170 loc) · 7.37 KB
/
cppn.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
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
import os
import sys
import argparse
import numpy as np
import torch
import tifffile
from torch import nn
from torch.nn import functional as F
from imageio import imwrite, imsave
# Because imageio uses the root logger instead of warnings package...
import logging
logging.getLogger().setLevel(logging.ERROR)
def load_args():
parser = argparse.ArgumentParser(description='cppn-pytorch')
parser.add_argument('--z', default=8, type=int, help='latent space width')
parser.add_argument('--n', default=1, type=int, help='images to generate')
parser.add_argument('--x_dim', default=2048, type=int, help='out image width')
parser.add_argument('--y_dim', default=2048, type=int, help='out image height')
parser.add_argument('--scale', default=10, type=float, help='mutiplier on z')
parser.add_argument('--c_dim', default=1, type=int, help='channels')
parser.add_argument('--net', default=32, type=int, help='net width')
parser.add_argument('--batch_size', default=1, type=int)
parser.add_argument('--interpolation', default=10, type=int)
parser.add_argument('--reinit', default=10, type=int, help='reinit generator every so often')
parser.add_argument('--exp', default='.', type=str, help='output fn')
parser.add_argument('--name_style', default='params', type=str, help='output fn')
parser.add_argument('--walk', action='store_true', help='interpolate')
parser.add_argument('--sample', action='store_true', help='sample n images')
args = parser.parse_args()
return args
class Generator(nn.Module):
def __init__(self, args):
super(Generator, self).__init__()
for k, v in vars(args).items():
setattr(self, k, v)
self.name = 'Generator'
dim = self.x_dim * self.y_dim * self.batch_size
self.linear_z = nn.Linear(self.z, self.net)
self.linear_x = nn.Linear(1, self.net, bias=False)
self.linear_y = nn.Linear(1, self.net, bias=False)
self.linear_r = nn.Linear(1, self.net, bias=False)
self.linear_h = nn.Linear(self.net, self.net)
self.linear_out = nn.Linear(self.net, self.c_dim)
self.sigmoid = nn.Sigmoid()
def forward(self, inputs):
x, y, z, r = inputs
n_points = self.x_dim * self.y_dim
ones = torch.ones(n_points, 1, dtype=torch.float)
z_scaled = z.view(self.batch_size, 1, self.z) * ones * self.scale
z_pt = self.linear_z(z_scaled.view(self.batch_size*n_points, self.z))
x_pt = self.linear_x(x.view(self.batch_size*n_points, -1))
y_pt = self.linear_y(y.view(self.batch_size*n_points, -1))
r_pt = self.linear_r(r.view(self.batch_size*n_points, -1))
U = z_pt + x_pt + y_pt + r_pt
H = torch.tanh(U)
H = F.elu(self.linear_h(H))
H = F.softplus(self.linear_h(H))
H = torch.tanh(self.linear_h(H))
x = .5 * torch.sin(self.linear_out(H)) + .5
img = x.reshape(self.batch_size, self.y_dim, self.x_dim, self.c_dim)
#print ('G out: ', img.shape)
return img
def coordinates(args):
x_dim, y_dim, scale = args.x_dim, args.y_dim, args.scale
n_points = x_dim * y_dim
x_range = scale*(np.arange(x_dim)-(x_dim-1)/2.0)/(x_dim-1)/0.5
y_range = scale*(np.arange(y_dim)-(y_dim-1)/2.0)/(y_dim-1)/0.5
x_mat = np.matmul(np.ones((y_dim, 1)), x_range.reshape((1, x_dim)))
y_mat = np.matmul(y_range.reshape((y_dim, 1)), np.ones((1, x_dim)))
r_mat = np.sqrt(x_mat*x_mat + y_mat*y_mat)
x_mat = np.tile(x_mat.flatten(), args.batch_size).reshape(args.batch_size, n_points, 1)
y_mat = np.tile(y_mat.flatten(), args.batch_size).reshape(args.batch_size, n_points, 1)
r_mat = np.tile(r_mat.flatten(), args.batch_size).reshape(args.batch_size, n_points, 1)
x_mat = torch.from_numpy(x_mat).float()
y_mat = torch.from_numpy(y_mat).float()
r_mat = torch.from_numpy(r_mat).float()
return x_mat, y_mat, r_mat
def sample(args, netG, z):
x_vec, y_vec, r_vec = coordinates(args)
image = netG((x_vec, y_vec, z, r_vec))
return image
def init(model):
for layer in model.modules():
if isinstance(layer, nn.Linear):
nn.init.normal_(layer.weight.data)
return model
def latent_walk(args, z1, z2, n_frames, netG):
delta = (z2 - z1) / (n_frames + 1)
total_frames = n_frames + 2
states = []
for i in range(total_frames):
z = z1 + delta * float(i)
if args.c_dim == 1:
states.append(sample(args, netG, z)[0]*255)
else:
states.append(sample(args, netG, z)[0]*255)
states = torch.stack(states).detach().numpy()
return states
def cppn(args):
seed = np.random.randint(123456789)
np.random.seed(seed)
torch.manual_seed(seed)
if not os.path.exists('./trials/'):
os.makedirs('./trials/')
subdir = args.exp
if not os.path.exists('trials/'+subdir):
os.makedirs('trials/'+subdir)
else:
while os.path.exists('trials/'+subdir):
response = input('Exp Directory Exists, rename (y/n/overwrite):\t')
if response == 'y':
subdir = input('New Exp Directory Name:\t')
elif response == 'overwrite':
break
else:
print ('Exiting...')
sys.exit(0)
os.makedirs('trials/'+subdir, exist_ok=True)
if args.name_style == 'simple':
suff = 'image'
if args.name_style == 'params':
suff = 'z-{}_scale-{}_cdim-{}_net-{}'.format(args.z, args.scale, args.c_dim, args.net)
netG = init(Generator(args))
print (netG)
n_images = args.n
zs = []
for _ in range(n_images):
zs.append(torch.zeros(1, args.z).uniform_(-1.0, 1.0))
if args.walk:
k = 0
for i in range(n_images):
if i+1 not in range(n_images):
images = latent_walk(args, zs[i], zs[0], args.interpolation, netG)
break
images = latent_walk(args, zs[i], zs[i+1], args.interpolation, netG)
for img in images:
save_fn = 'trials/{}/{}_{}'.format(subdir, suff, k)
print ('saving PNG image at: {}'.format(save_fn))
imwrite(save_fn+'.png', img)
k += 1
print ('walked {}/{}'.format(i+1, n_images))
elif args.sample:
zs, _ = torch.stack(zs).sort()
for i, z in enumerate(zs):
img = sample(args, netG, z).cpu().detach().numpy()
if args.c_dim == 1:
img = img[0]
else:
img = img[0]
img = img * 255
metadata = dict(seed=str(seed),
z_sample=str(list(z.numpy()[0])),
z=str(args.z),
c_dim=str(args.c_dim),
scale=str(args.scale),
net=str(args.net))
save_fn = 'trials/{}/{}_{}'.format(subdir, suff, i)
print ('saving TIFF/PNG image pair at: {}'.format(save_fn))
tifffile.imsave(save_fn+'.tif',
img.astype('u1'),
metadata=metadata)
imwrite(save_fn+'.png'.format(subdir, suff, i), img)
else:
print ('No action selected. Exiting...')
print ('If this is an error, check command line arguments for ' \
'generating images')
sys.exit(0)
if __name__ == '__main__':
args = load_args()
cppn(args)