-
Notifications
You must be signed in to change notification settings - Fork 4
/
utils_gans.py
274 lines (226 loc) · 9.99 KB
/
utils_gans.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
269
270
271
272
273
274
import numpy as np
import matplotlib.pyplot as plt
import os
import sys
import torch
from torch import optim
from torchvision import transforms
import pandas as pd
from sklearn.model_selection import train_test_split
sys.path.append('/workspace/stylegan2-pytorch')
from sequencedataloader import txt_dataloader_styleGAN
from torchvision import transforms, utils
import random
# Creating seeds to make results reproducible
def seed_everything(seed_value):
np.random.seed(seed_value)
random.seed(seed_value)
torch.manual_seed(seed_value)
os.environ['PYTHONHASHSEED'] = str(seed_value)
if torch.cuda.is_available():
torch.cuda.manual_seed(seed_value)
torch.cuda.manual_seed_all(seed_value)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = True
seed = 2022
seed_everything(seed)
# Handle data
gantransform = transforms.Compose(
[
transforms.Resize((256, 256)),
transforms.ToTensor(),
transforms.Normalize(
(0.5, 0.5, 0.5), (0.5, 0.5, 0.5), inplace=True),
]
)
def load_isic_gandata(data_path='/workspace/data/data/melanoma_external_256/'):
# ISIC Dataset
df = pd.read_csv(os.path.join(data_path, "train_concat.csv"))
train_img_dir = os.path.join(data_path, "train/train/")
df['image_name'] = [os.path.join(
train_img_dir, df.iloc[index]['image_name'] + '.jpg') for index in range(len(df))]
train_split, valid_split = train_test_split(
df, stratify=df.target, test_size=0.20, random_state=42)
train_df=pd.DataFrame(train_split)
validation_df=pd.DataFrame(valid_split)
training_dataset = CustomGanDataset(df=train_df, transform=gantransform)
testing_dataset = CustomGanDataset(df=validation_df, transform=gantransform)
num_examples = {"trainset":len(training_dataset), "testset":len(testing_dataset)}
return training_dataset, testing_dataset, num_examples
def load_isic_by_patient_client(
partition,
data_path='/workspace/data/data/melanoma_external_256'
):
# Load data
df = pd.read_csv(os.path.join(data_path, "train_concat.csv"))
train_img_dir = os.path.join(data_path, "train/train/")
df['image_name'] = [os.path.join(
train_img_dir, df.iloc[index]['image_name'] + '.jpg'
) for index in range(len(df))]
df["patient_id"] = df["patient_id"].fillna('nan')
# Split by Patient
patient_groups = df.groupby('patient_id') #37311
melanoma_groups_list = [
patient_groups.get_group(x) for x in patient_groups.groups if patient_groups.get_group(x)['target'].unique().all()==1] # 4188 - after adding na 4525
benign_groups_list = [
patient_groups.get_group(x) for x in patient_groups.groups if 0 in patient_groups.get_group(x)['target'].unique()] # 2055 - 33123
np.random.shuffle(melanoma_groups_list)
np.random.shuffle(benign_groups_list)
if partition == 2:
df_b_test = pd.concat(benign_groups_list[1800:]) # 4462
df_b_train = pd.concat(benign_groups_list[800:1800]) # 16033 - TOTAL 20495 samples
df_m_test = pd.concat(melanoma_groups_list[170:281]) # 340
df_m_train = pd.concat(melanoma_groups_list[281:800]) # 1970 - TOTAL: 2310 samples
elif partition == 1:
df_b_test = pd.concat(benign_groups_list[130:250]) # 1949
df_b_train = pd.concat(benign_groups_list[250:800]) # 8609 - TOTAL 10558 samples
df_m_test = pd.concat(melanoma_groups_list[1230:]) # 303
df_m_train = pd.concat(melanoma_groups_list[800:1230]) # 1407 - TOTAL 1710 samples
else:
df_b_test = pd.concat(benign_groups_list[:30]) # 519
df_b_train = pd.concat(benign_groups_list[30:130]) # 1551 - TOTAL: 2070 samples
df_m_test = pd.concat(melanoma_groups_list[:70]) # 191
df_m_train = pd.concat(melanoma_groups_list[70:170]) # 314 - TOTAL: 505 samples
train_split = pd.concat([df_b_train, df_m_train])
valid_split = pd.concat([df_b_test, df_m_test])
train_df=pd.DataFrame(train_split)
validation_df=pd.DataFrame(valid_split)
training_dataset = CustomGanDataset(df=train_df, transform=gantransform)
testing_dataset = CustomGanDataset(df=validation_df, transform=gantransform)
num_examples = {"trainset" : len(training_dataset), "testset" : len(testing_dataset)}
return training_dataset, testing_dataset, num_examples
class CustomGanDataset(txt_dataloader_styleGAN):
def __init__(self, df: pd.DataFrame, conditional=True, transform=None,
usePIL=True, isSequence=False, GANflag = False, verbose=True):
self.df = df
self.transform = transform
self.conditional = conditional
self.GANflag = GANflag
self.verbose = verbose
self.images = list(df["image_name"])
self.labels = list(df["target"].values)
if self.verbose:
print('Images loaded: ' + str(len(self.images)) + '\n')
self.usePIL = usePIL
self.isSequence = isSequence
self.imgs = list(zip(self.images, self.labels))
def load_experiment_partition2(trainset, testset, num_examples, idx):
"""Load 1/5th of the training and test data to simulate a partition."""
assert idx in range(3)
if idx==0:
train_partition = torch.utils.data.Subset(
trainset, range(0, 2000)
)
test_partition = torch.utils.data.Subset(
testset, range(0,502)
)
print('Train loaded: ' + str(len(train_partition)) + '\n')
print('Test loaded: ' + str(len(test_partition)) + '\n')
train_partition, test_partition = trainset.images[0:2000], testset.images[0:502]
elif idx==1:
train_partition = torch.utils.data.Subset(
trainset, range(5000, 10000)
)
test_partition = torch.utils.data.Subset(
testset, range(600, 1855)
)
print('Train loaded: ' + str(len(train_partition)) + '\n')
print('Test loaded: ' + str(len(test_partition)) + '\n')
train_partition = trainset.images[5000:10000]
test_partition = testset.images[600:1855]
else:
train_partition = torch.utils.data.Subset(
trainset, range(10000, 20000)
)
test_partition = torch.utils.data.Subset(
testset, range(2000, 4510)
)
print('Train loaded: ' + str(len(train_partition)) + '\n')
print('Test loaded: ' + str(len(test_partition)) + '\n')
train_partition = trainset.images[10000:20000]
test_partition = testset.images[2000:4510]
num_examples = {"trainset" : len(train_partition), "testset" : len(test_partition)}
return (train_partition, test_partition, num_examples)
# Handle model
def load_ganmodel(args):
if args.arch == 'stylegan2':
from model_conditional import Generator, Discriminator
elif args.arch == 'swagan':
from swagan_conditional import Generator, Discriminator
from train_conditional import accumulate
generator = Generator(
args.size, args.latent, args.n_mlp,
num_classes=args.num_classes,
channel_multiplier=args.channel_multiplier
)
discriminator = Discriminator(
args.size, channel_multiplier=args.channel_multiplier,
num_classes=args.num_classes
)
g_ema = Generator(
args.size, args.latent, args.n_mlp, num_classes=args.num_classes, channel_multiplier=args.channel_multiplier
)
g_ema.eval()
accumulate(g_ema, generator, 0)
g_reg_ratio = args.g_reg_every / (args.g_reg_every + 1)
d_reg_ratio = args.d_reg_every / (args.d_reg_every + 1)
g_optim = optim.Adam(
generator.parameters(),
lr=args.lr * g_reg_ratio,
betas=(0 ** g_reg_ratio, 0.99 ** g_reg_ratio),
)
d_optim = optim.Adam(
discriminator.parameters(),
lr=args.lr * d_reg_ratio,
betas=(0 ** d_reg_ratio, 0.99 ** d_reg_ratio),
)
return generator, discriminator, g_ema, g_optim, d_optim
def train_gan(args, loader, generator,
discriminator, g_ema,
g_optim, d_optim,
device, id):
from train_conditional import train
# Training model
generator.to(device)
discriminator.to(device)
g_ema.to(device)
print('Starts training...')
return train(args, loader, generator, discriminator,
g_optim, d_optim, g_ema,
device, args.partition, id)
def val_gan(args, g_ema, g_module, d_module,
g_optim, d_optim, i, clin_id=0,
save_path="/workspace/data/sample/test_client"):
g_ema.to(args.device)
g_ema.eval()
sample_z = torch.randn(args.n_sample, args.latent, device=args.device)
sample_labels = torch.tensor([0,1]).repeat(args.n_sample//args.num_classes)
sample_labels = torch.nn.functional.one_hot(sample_labels, num_classes=args.num_classes).float().to(args.device)
# Turning off gradients for validation, saves memory and computations
with torch.no_grad():
sample, _ = g_ema([sample_z], sample_labels)
utils.save_image(
sample,
f"{save_path}-{str(clin_id)}/{str(i).zfill(6)}.png",
nrow=int(args.n_sample ** 0.5),
normalize=True,
range=(-1, 1),
)
grid = utils.make_grid(sample, nrow=int(
args.n_sample ** 0.5),normalize=True, range=(-1, 1))
# Add 0.5 after unnormalizing to [0, 255] to round to nearest integer
ndarr = grid.mul(255).add_(0.5).clamp_(0, 255).permute(
1, 2, 0).to('cpu', torch.uint8).numpy().astype(np.float32)
im = ndarr
torch.save(
{
"g": g_module.state_dict(),
"d": d_module.state_dict(),
"g_ema": g_ema.state_dict(),
"g_optim": g_optim.state_dict(),
"d_optim": d_optim.state_dict(),
"args": args,
},
f"{save_path}-{str(clin_id)}/{str(i).zfill(6)}.pt",
)
return im, f"Iter:{str(i).zfill(6)}"