-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrain.py
137 lines (111 loc) · 4.24 KB
/
train.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
from model import TemporalGenerator, ImageGenerator, Discriminator_E
from loader import GIF
import os
import imageio
import numpy as np
import torch
from torch.utils.data import DataLoader
from torch.optim import Adam
from torch.autograd import Variable
def reset_grad(optims):
for optim in optims:
optim.zero_grad()
def to_gif(vs, name, how_many=2):
# var = N x C x D x H x W
vs = vs.cpu().data.numpy()
vs = np.einsum('ijklm->iklmj', vs)
os.makedirs('outputs', exist_ok=True)
for i, v in enumerate(vs):
v = (v + 1) * 127.5
v = v.astype(np.uint8)
imageio.mimsave(f'outputs/{epoch}_{batch_idx}_{i}_{name}.gif', v)
if i == (how_many - 1): break
if __name__ == '__main__':
batch_size = 16
G_TG = TemporalGenerator().cuda()
G_IG = ImageGenerator(3).cuda()
D_E = Discriminator_E(3).cuda()
D_TG = TemporalGenerator().cuda()
D_IG = ImageGenerator(3).cuda()
def D(X):
DE = D_E(X)
DTG = D_TG(DE)
DIG = D_IG(DE, DTG)
X_recon = DIG
return torch.mean(torch.sum(torch.abs(X - X_recon), 1))
mm_dataset = GIF()
start_epoch = 0
lr = 1e-3
optim_G_TG = Adam(G_TG.parameters(), lr=lr)
optim_G_IG = Adam(G_IG.parameters(), lr=lr)
optim_D_E = Adam(D_E.parameters(), lr=lr)
optim_D_TG = Adam(D_TG.parameters(), lr=lr)
optim_D_IG = Adam(D_IG.parameters(), lr=lr)
optims = [optim_G_TG, optim_G_IG, optim_D_E, optim_D_TG, optim_D_IG]
if os.path.isfile('checkpoint.pth'):
print("=> loading checkpoint")
checkpoint = torch.load('checkpoint.pth')
start_epoch = checkpoint['epoch']
G_TG.load_state_dict(checkpoint['G_TG'])
G_IG.load_state_dict(checkpoint['G_IG'])
D_E.load_state_dict(checkpoint['D_E'])
D_TG.load_state_dict(checkpoint['D_TG'])
D_IG.load_state_dict(checkpoint['D_IG'])
optim_G_TG.load_state_dict(checkpoint['optim_G_TG'])
optim_G_IG.load_state_dict(checkpoint['optim_G_IG'])
optim_D_E.load_state_dict(checkpoint['optim_D_E'])
optim_D_TG.load_state_dict(checkpoint['optim_D_TG'])
optim_D_IG.load_state_dict(checkpoint['optim_D_IG'])
for epoch in range(start_epoch, 1000000):
mm_loader = DataLoader(mm_dataset, batch_size=batch_size, shuffle=True)
k = 0
lam = 1e-3
gamma = 0.5
for batch_idx, data in enumerate(mm_loader):
# Sample Data
X = Variable(data.cuda())
# Discriminator
z0_D = Variable(torch.rand(batch_size, 100, 1).cuda())
z1_D = G_TG(z0_D)
Fake_D = G_IG(z0_D, z1_D)
D_X = D(X)
D_Fake_D = D(Fake_D)
L_D = D_X - k * D_Fake_D
L_D.backward()
optim_D_E.step()
optim_D_TG.step()
optim_D_IG.step()
reset_grad(optims)
# Generator
z0_G = Variable(torch.rand(batch_size, 100, 1).cuda())
z1_G = G_TG(z0_G)
Fake_G = G_IG(z0_G, z1_G)
D_Fake_G = D(Fake_G)
L_G = D_Fake_G
L_G.backward()
optim_G_TG.step()
optim_G_IG.step()
reset_grad(optims)
# Update k, the equlibrium
k = k + lam * (gamma * D_X - D_Fake_G)
k = k.data[0] # Dismiss Variable
measure = D_X + torch.abs(gamma * D_X - D_Fake_G)
print(f'Epoch-{epoch}, Batch-{batch_idx}, Convergence measure: {measure.data[0]:.4}')
if batch_idx % 100 == 0:
to_gif(Fake_G, 'fake_g')
to_gif(D_IG(D_E(Fake_G), D_TG(D_E(Fake_G))), 'fake_g_autoencoded')
to_gif(X, 'real')
to_gif(D_IG(D_E(X), D_TG(D_E(X))), 'real_autoencoded')
torch.save({
'epoch': epoch,
'G_TG': G_TG.state_dict(),
'G_IG': G_IG.state_dict(),
'D_E': D_E.state_dict(),
'D_TG': D_TG.state_dict(),
'D_IG': D_IG.state_dict(),
'optim_G_TG': optim_G_TG.state_dict(),
'optim_G_IG': optim_G_IG.state_dict(),
'optim_D_E': optim_D_E.state_dict(),
'optim_D_TG': optim_D_TG.state_dict(),
'optim_D_IG': optim_D_IG.state_dict(),
}, 'checkpoint.pth')