-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainer.py
138 lines (113 loc) · 5.18 KB
/
trainer.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
import torch
import matplotlib.pyplot as plt
from torchvision.utils import make_grid
from tqdm.notebook import tqdm
from IPython.display import clear_output
######################################
####### Function to show images #####
######################################
def showImage(img_data, title):
'''
Function for visualizing images: Given a tensor of images, number of images, and
size per image, plots and prints the images in an uniform grid.
'''
img_data = ((img_data + 1.)/2).detach().cpu()
img_grid = make_grid(img_data[:4], nrow=4)
plt.axis('off')
plt.title(title)
plt.imshow(img_grid.permute(1, 2, 0).squeeze())
plt.show()
######################################
########### Predictor ############
######################################
class Predictor():
def __init__(self, model, transform):
self.model=model.eval()
self.transform=transform
def predict(self, dataloader, path='test'):
names = dataloader.dataset.images
for i, img in enumerate(tqdm(dataloader)):
hd_img_data = self.model(img)
hd_img = self.transform(hd_img_data)
hd_img.save(path+'/hd_'+names[i])
######################################
######## Generator`s WarmUp #######
######################################
class GeneratorWarmUp():
def __init__(self, generator, criterion, optimizer, device='cuda:0'):
self.generator=generator.to(device)
self.criterion=criterion.to(device)
self.optimizer=optimizer
def start(self, dataloader, epochs=500, device='cuda:0', name='SRResNet'):
self.losses=[]
for epoch in range(epochs):
loss = 0.
for x, real in tqdm(dataloader):
x = x.to(device)
real = real.to(device)
with torch.cuda.amp.autocast(enabled=(device=='cuda:0')):
fake = self.generator(x)
loss = self.criterion(fake, real)
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
loss += loss.item()
self.losses.append(loss/len(dataloader))
clear_output(wait=True)
print('Epoch {}: Generator`s ({}) loss: {:.5f}'.format(epoch+1, name, self.losses[-1]))
showImage(2*x - 1., 'Input Low-Resolution Image')
showImage(fake.to(real.dtype), 'Generated High-Resolution Image')
showImage(real, 'Real High-Resolution Image')
######################################
###### Super-Resolution Trainer #####
######################################
class Trainer():
def __init__(self,
generator, discriminator,
g_criterion, d_criterion,
g_optimizer, d_optimizer, device='cuda:0'):
self.generator = generator.to(device)
self.discriminator = discriminator.to(device)
self.g_criterion = g_criterion.to(device)
self.d_criterion = d_criterion.to(device)
self.g_optimizer = g_optimizer
self.d_optimizer = d_optimizer
def fit(self, dataloader, epochs=500, device='cuda:0'):
g_losses=[]
d_losses=[]
for epoch in range(epochs):
ge_loss=0.
de_loss=0.
for x, real in tqdm(dataloader):
x = x.to(device)
real = real.to(device)
with torch.cuda.amp.autocast(enabled=(device=='cuda:0')):
# Generator`s loss
fake = self.generator(x)
fake_pred = self.discriminator(fake)
g_loss = self.g_criterion(fake_pred, fake, real)
# Discriminator`s loss
fake = self.generator(x).detach()
fake_pred = self.discriminator(fake)
real_pred = self.discriminator(real)
d_loss = self.d_criterion(fake_pred, real_pred)
# Generator`s params update
self.g_optimizer.zero_grad()
g_loss.backward()
self.g_optimizer.step()
# Discriminator`s params update
self.d_optimizer.zero_grad()
d_loss.backward()
self.d_optimizer.step()
ge_loss += g_loss.item()
de_loss += d_loss.item()
g_losses.append(ge_loss/len(dataloader))
d_losses.append(de_loss/len(dataloader))
clear_output(wait=True)
print(f'::::::::::::::::: Epoch {epoch+1} :::::::::::::::::')
print(f'::::::::::: Generator loss: {g_losses[-1]:.3f} :::::::::::')
print(f'::::::::: Discriminator loss: {d_losses[-1]:.3f} :::::::::')
showImage(2*x - 1., 'Input Low-Resolution Image')
showImage(fake.to(real.dtype), 'Generated High-Resolution Image')
showImage(real, 'Real High-Resolution Image')
self.data={'g_loss': g_losses, 'd_loss': d_losses}