-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
160 lines (129 loc) · 6.46 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
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
import torch
from torch.utils.data import DataLoader
import numpy as np
import argparse
import yaml
from tqdm import tqdm
import os
from glob import glob
import gc
from model.generator import Generator
from model.discriminator import Discriminator
from utils.r1_loss import generator_loss, discriminator_loss
from dataset import Anime
from utils.history import History
class Trainer:
def __init__(self, hp):
self.device = torch.device('cuda')
self.generator = Generator(
res = hp['generator']['resolution'],
latent_size = hp['generator']['latent_size'],
deep_mapping = hp['generator']['deep_mapping']
).to(self.device)
self.optim_G = getattr(torch.optim, hp['generator']['optim'])(
params = self.generator.parameters(),
alpha = hp['generator']['alpha'],
lr = hp['generator']['learning_rate']
)
if hp['generator']['loss'] == 'r1':
self.generator_loss = generator_loss
self.z_dim = hp['generator']['latent_size']
self.discriminator = Discriminator(hp['discriminator']['resolution']).to(self.device)
self.optim_D = getattr(torch.optim, hp['discriminator']['optim'])(
params = self.discriminator.parameters(),
alpha = hp['discriminator']['alpha'],
lr = hp['discriminator']['learning_rate']
)
if hp['discriminator']['loss'] == 'r1':
self.discriminator_loss = discriminator_loss
self.lambda_gp = hp['discriminator']['lambda_gp']
anime = Anime(hp['dataset_dir'], hp['image_size'])
self.dataloader = DataLoader(anime, batch_size=hp['batch_size'], shuffle=True, pin_memory=False, drop_last=True)
self.start_epoch = 0
self.epochs = hp['epochs']
self.batch_size = hp['batch_size']
self.history = History()
self.weight_dir = hp['weight_dir']
self.Z = torch.randn((hp['generate_no_image']*hp['generate_no_image'], self.z_dim), device=self.device)
if os.path.exists(self.weight_dir):
self.load_model()
else:
os.mkdir(hp['weight_dir'])
self.images_dir = hp['images_dir']
if not os.path.exists(self.images_dir):
os.mkdir(self.images_dir)
self.save_frequency = hp['save_frequency']
self.images_dir = hp['images_dir']
self.generate_no_image = hp['generate_no_image']
def generate_images(self, count=1):
z = torch.randn((count, self.z_dim), device=self.device)
return self.generator(z)
def train_generator(self, fake_images):
logits_fake = self.discriminator(fake_images)
return self.generator_loss(logits_fake)
def train_discriminator(self, real_images, fake_images):
return self.discriminator_loss(self.discriminator, real_images, fake_images, self.lambda_gp)
def train(self):
print("Training StyleGAN2")
print("Generator : ")
print(self.generator)
print("Discriminator : ")
print(self.discriminator)
for epoch in range(self.start_epoch, self.epochs):
g_losses, d_losses = list(), list()
for real_images in tqdm(self.dataloader):
self.generator.zero_grad()
real_images = real_images.to(self.device)
fake_images = self.generate_images(self.batch_size).detach()
d_loss = self.train_discriminator(real_images, fake_images)
self.discriminator.zero_grad()
d_loss.backward()
self.optim_D.step()
d_losses.append(d_loss)
fake_images = self.generate_images(self.batch_size)
g_loss = self.train_generator(fake_images)
self.generator.zero_grad()
g_loss.backward()
self.optim_G.step()
g_losses.append(g_loss)
self.history.save_history(epoch, sum(g_losses)/len(g_losses), sum(d_losses)/len(d_losses))
if epoch%self.save_frequency==0:
self.save_model(epoch)
self.history.plot_images(self.generator(self.Z).cpu(), epoch, self.images_dir, self.generate_no_image)
gc.collect()
torch.cuda.empty_cache()
self.history.plot_loss(self.images_dir)
def save_model(self, epoch):
pretrained_models = glob(os.path.join(self.weight_dir, 'model_*.pt'))
if len(pretrained_models) >= 5:
earliest_model = min(pretrained_models, key=lambda x: int(x.split('_')[-1].split('.')[0]))
os.remove(earliest_model)
torch.save(self.generator.state_dict(), os.path.join(self.weight_dir, f'model_{epoch}.pt'))
training_state = dict()
training_state['generator'] = self.generator.state_dict()
training_state['optim_G'] = self.optim_G.state_dict()
training_state['discriminator'] = self.discriminator.state_dict()
training_state['optim_D'] = self.optim_D.state_dict()
training_state['epoch'] = epoch
training_state['history'] = self.history
training_state['Z'] = self.Z
torch.save(training_state, os.path.join(self.weight_dir, 'training_state.pt'))
def load_model(self):
if not os.path.exists(os.path.join(self.weight_dir, 'training_state.pt')):
return
print("Resuming training from previous epoch")
training_state = torch.load(os.path.join(self.weight_dir, 'training_state.pt'))
self.generator.load_state_dict(training_state['generator'])
self.optim_G.load_state_dict(training_state['optim_G'])
self.discriminator.load_state_dict(training_state['discriminator'])
self.optim_D.load_state_dict(training_state['optim_D'])
self.start_epoch = training_state['epoch']+1
self.history = training_state['history']
self.Z = training_state['Z']
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--hyperparameters', type=str, default='hyperparameters.yaml', help='The path for hyperparameters.yaml file')
args = parser.parse_args()
hyperparameters = yaml.safe_load(open(args.hyperparameters, 'r'))
trainer = Trainer(hyperparameters)
trainer.train()