-
Notifications
You must be signed in to change notification settings - Fork 1
/
train_golf.py
268 lines (206 loc) · 9.05 KB
/
train_golf.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
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
"""
videogan PT version
Original torch ver from https://github.com/cvondrick/videogan
TF ver from https://github.com/GV1028/videogan
"""
import torch
import torch.nn as nn
import torch.nn.init as init
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
import os
import scipy.misc
import numpy as np
import glob
from utils import *
import sys
from argparse import ArgumentParser
from datetime import datetime
from random import shuffle
from skimage.transform import resize
parser = ArgumentParser()
parser.add_argument(
"-e", help="Epoch to train",
type=int, default=10
)
parser.add_argument(
"-d", help="The dimension of each video, must be of shape [3,32,64,64]",
nargs='*', default=[3,32,64,64]
)
parser.add_argument(
"-zd", help="The dimension of latent vector [100]",
type=int, default=100
)
parser.add_argument(
"-nb", help="The size of batch images [64]",
type=int, default=20
)
parser.add_argument(
"-l", help="The value of sparsity regularizer [0.1]",
type=float, default=0.1
)
parser.add_argument(
"-c", help="The checkpoint file name",
type=str, default="2019-04-04-01-09-10_1_4714"
)
parser.add_argument(
"-s", help="Saving checkpoint file, every [1] epochs",
type=int, default=1
)
args = parser.parse_args()
class Generator(torch.nn.Module):
def __init__(self, zdim=args.zd):
super(Generator, self).__init__()
self.zdim = zdim
# Background
self.conv1b = nn.ConvTranspose2d(zdim, 512, [4,4], [1,1])
self.bn1b = nn.BatchNorm2d(512)
self.conv2b = nn.ConvTranspose2d(512, 256, [4,4], [2,2], [1,1])
self.bn2b = nn.BatchNorm2d(256)
self.conv3b = nn.ConvTranspose2d(256, 128, [4,4], [2,2], [1,1])
self.bn3b = nn.BatchNorm2d(128)
self.conv4b = nn.ConvTranspose2d(128, 64, [4,4], [2,2], [1,1])
self.bn4b = nn.BatchNorm2d(64)
self.conv5b = nn.ConvTranspose2d(64, 3, [4,4], [2,2], [1,1])
# Foreground
self.conv1 = nn.ConvTranspose3d(zdim, 512, [2,4,4], [1,1,1])
self.bn1 = nn.BatchNorm3d(512)
self.conv2 = nn.ConvTranspose3d(512, 256, [4,4,4], [2,2,2], [1,1,1])
self.bn2 = nn.BatchNorm3d(256)
self.conv3 = nn.ConvTranspose3d(256, 128, [4,4,4], [2,2,2], [1,1,1])
self.bn3 = nn.BatchNorm3d(128)
self.conv4 = nn.ConvTranspose3d(128, 64, [4,4,4], [2,2,2], [1,1,1])
self.bn4 = nn.BatchNorm3d(64)
self.conv5 = nn.ConvTranspose3d(64, 3, [4,4,4], [2,2,2], [1,1,1])
# Mask
self.conv5m = nn.ConvTranspose3d(64, 1, [4,4,4], [2,2,2], [1,1,1])
# Init weights
for m in self.modules():
classname = m.__class__.__name__
if classname.lower().find('conv') != -1:
nn.init.normal_(m.weight, mean=0, std=0.01)
nn.init.constant_(m.bias, 0)
elif classname.lower().find('bn') != -1:
nn.init.normal_(m.weight, mean=1, std=0.02)
nn.init.constant_(m.bias, 0)
def forward(self, z):
# Background
b = F.relu(self.bn1b(self.conv1b(z.unsqueeze(2).unsqueeze(3))))
b = F.relu(self.bn2b(self.conv2b(b)))
b = F.relu(self.bn3b(self.conv3b(b)))
b = F.relu(self.bn4b(self.conv4b(b)))
b = torch.tanh(self.conv5b(b)).unsqueeze(2) # b, 3, 1, 64, 64
# Foreground
f = F.relu(self.bn1(self.conv1(z.unsqueeze(2).unsqueeze(3).unsqueeze(4))))
f = F.relu(self.bn2(self.conv2(f)))
f = F.relu(self.bn3(self.conv3(f)))
f = F.relu(self.bn4(self.conv4(f)))
m = torch.sigmoid(self.conv5m(f)) # b, 1, 32, 64, 64
f = torch.tanh(self.conv5(f)) # b, 3, 32, 64, 64
out = m*f + (1-m)*b
return out, f, b, m
class Discriminator(torch.nn.Module):
def __init__(self):
super(Discriminator, self).__init__()
self.conv1 = nn.Conv3d(3, 64, [4,4,4], [2,2,2], [1,1,1])
self.conv2 = nn.Conv3d(64, 128, [4,4,4], [2,2,2], [1,1,1])
self.bn2 = nn.BatchNorm3d(128)
self.conv3 = nn.Conv3d(128, 256, [4,4,4], [2,2,2], [1,1,1])
self.bn3 = nn.BatchNorm3d(256)
self.conv4 = nn.Conv3d(256, 512, [4,4,4], [2,2,2], [1,1,1])
self.bn4 = nn.BatchNorm3d(512)
self.conv5 = nn.Conv3d(512, 1, [2,4,4], [1,1,1])
# Init weights
for m in self.modules():
classname = m.__class__.__name__
if classname.lower().find('conv') != -1:
nn.init.normal_(m.weight, mean=0, std=0.01)
nn.init.constant_(m.bias, 0)
elif classname.lower().find('bn') != -1:
nn.init.normal_(m.weight, mean=1, std=0.02)
nn.init.constant_(m.bias, 0)
def forward(self, x):
x = F.leaky_relu(self.conv1(x), 0.2)
x = F.leaky_relu(self.bn2(self.conv2(x)), 0.2)
x = F.leaky_relu(self.bn3(self.conv3(x)), 0.2)
x = F.leaky_relu(self.bn4(self.conv4(x)), 0.2)
x = self.conv5(x)
return x
def main():
if not os.path.exists("./checkpoints"):
os.makedirs("./checkpoints")
if not os.path.exists("./genvideos"):
os.makedirs("./genvideos")
# Model def
G = Generator(zdim=args.zd).cuda()
D = Discriminator().cuda()
# Optimizer
params_G = list(filter(lambda p: p.requires_grad, G.parameters()))
optimizer_G = optim.Adam(params_G, lr=0.0002, betas=(0.5,0.999))
params_D = list(filter(lambda p: p.requires_grad, D.parameters()))
optimizer_D = optim.Adam(params_D, lr=0.0002, betas=(0.5,0.999))
# Load pretrained
if args.c is not None:
G.load_state_dict(torch.load("./checkpoints/{}_G.pth".format(args.c)).state_dict(), strict=True)
D.load_state_dict(torch.load("./checkpoints/{}_D.pth".format(args.c)).state_dict(), strict=True)
print("Model restored")
# Train
# data file list
f = open("./golf/golf.txt", 'r')
data_lists = f.readlines()
shuffle(data_lists)
for epoch in range(args.e):
for counter in range(int(len(data_lists)/args.nb)):
# Data
real_video = []
i = 0
while True:
b = data_lists[counter*args.nb + i]
i += 1
img = np.asarray(PIL.Image.open(("./golf/frames-stable-many/{}").format(b.strip()))) / 127.5 - 1
frames = []
if img.shape[0] < 128*args.d[1]:
continue
for f in range(args.d[1]):
frames.append( resize(img[f*128:(f+1)*128], (64,64), anti_aliasing=True) )
real_video.append( np.stack(frames, 0) )
if len(real_video) >= args.nb:
break
real_video = Variable(torch.from_numpy(np.stack(real_video, 0).astype(np.float32).transpose((0,4,1,2,3))), requires_grad=True).cuda()
# D
noise = torch.from_numpy(np.random.normal(0, 1, size=[args.nb,args.zd]).astype(np.float32)).cuda()
with torch.no_grad():
fake_video, f, b, m = G(noise)
logit_real = D(real_video)
logit_fake = D(fake_video.detach())
prob_real = torch.mean(torch.sigmoid(logit_real))
prob_fake = torch.mean(torch.sigmoid(logit_fake))
loss_real = F.binary_cross_entropy_with_logits(logit_real, torch.ones_like(logit_real))
loss_fake = F.binary_cross_entropy_with_logits(logit_fake, torch.zeros_like(logit_fake))
loss_D = torch.mean(torch.stack([loss_real, loss_fake]))
optimizer_D.zero_grad()
loss_D.backward()
optimizer_D.step()
# G
noise = torch.from_numpy(np.random.normal(0, 1, size=[args.nb,args.zd]).astype(np.float32)).cuda()
gen_video, f, b, m = G(noise)
logit_gen = D(gen_video)
loss_gen = F.binary_cross_entropy_with_logits(logit_gen, torch.ones_like(logit_gen))
loss_G = torch.mean(torch.stack([loss_gen])) + args.l*F.l1_loss(m, torch.zeros_like(m), True, True)
optimizer_G.zero_grad()
loss_G.backward()
optimizer_G.step()
# Print status
print("Epoch {:d}/{:d} | Iter {:d}/{:d} | D {:.4e} | G {:.4e} | Real {:.2f} | Fake {:.2f}".format(epoch, args.e, counter, int(len(data_lists)/args.nb), loss_D, loss_G, prob_real, prob_fake))
process_and_write_video(gen_video[0:1].cpu().data.numpy(), "curr_video")
process_and_write_image(b.cpu().data.numpy(), "curr_bg")
if (epoch+1) % args.s == 0:
process_and_write_video(gen_video[0:1].cpu().data.numpy(), "epoch{}_iter{}_video".format(epoch, counter))
process_and_write_image(b.cpu().data.numpy(), "epoch{}_iter{}_bg".format(epoch, counter))
curr_time = datetime.now().strftime('%Y-%m-%d-%H-%M-%S')
torch.save(G, "./checkpoints/{}_{}_{}_G.pth".format(curr_time, epoch, counter))
torch.save(D, "./checkpoints/{}_{}_{}_D.pth".format(curr_time, epoch, counter))
print ("Checkpoints saved")
if __name__ == '__main__':
main()