-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
388 lines (331 loc) · 18.7 KB
/
utils.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
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
"""
Copyright (C) 2018 NVIDIA Corporation. All rights reserved.
Licensed under the CC BY-NC-SA 4.0 license (https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode).
"""
import torch
#from torch.utils.serialization import load_lua
from torch.utils.data import DataLoader
from networks import Vgg16
from torch.autograd import Variable
from torch.optim import lr_scheduler
from torchvision import transforms
from data import ImageFilelist, ImageFolder, ImageFolder_Double
import torch.nn as nn
import os
import math
import torchvision.utils as vutils
import yaml
import numpy as np
import torch.nn.init as init
import time
import random
# Methods
# get_all_data_loaders : primary data loader interface (load trainA, testA, trainB, testB)
# get_data_loader_list : list-based data loader
# get_data_loader_folder : folder-based data loader
# get_config : load yaml file
# eformat :
# write_2images : save output image
# prepare_sub_folder : create checkpoints and images folders for saving outputs
# write_one_row_html : write one row of the html file for output images
# write_html : create the html file.
# write_loss
# slerp
# get_slerp_interp
# get_model_list
# load_vgg16
# load_inception
# vgg_preprocess
# get_scheduler
# weights_init
def get_all_data_loaders(conf):
batch_size = conf['batch_size']
num_workers = conf['num_workers']
if 'new_size' in conf:
new_size_a = new_size_b = conf['new_size']
else:
new_size_a = conf['new_size_a']
new_size_b = conf['new_size_b']
height = conf['crop_image_height']
width = conf['crop_image_width']
continuize_transforms = conf['continuize_transforms']
train_loader_ab = get_data_loader_folder_double(os.path.join(conf['data_root'], 'trainA'), os.path.join(conf['data_root'], 'trainB'),
False, batch_size, False, new_size_a, False, num_workers, False, continuize_transforms=continuize_transforms)
train_loader_aTbT = get_data_loader_folder_double(os.path.join(conf['data_root'], 'trainA'), os.path.join(conf['data_root'], 'trainB'),
True, batch_size, False, new_size_a, False, num_workers, False, continuize_transforms=continuize_transforms)
test_loader_ab = get_data_loader_folder_double(os.path.join(conf['data_root'], 'testA'), os.path.join(conf['data_root'], 'testB'),
False, batch_size, False, new_size_a, False, num_workers, False, continuize_transforms=continuize_transforms)
test_loader_aTbT = get_data_loader_folder_double(os.path.join(conf['data_root'], 'testA'), os.path.join(conf['data_root'], 'testB'),
True, batch_size, False, new_size_a, False, num_workers, False, continuize_transforms=continuize_transforms)
return train_loader_ab, train_loader_aTbT, test_loader_ab, test_loader_aTbT
def get_data_loader_folder_double(input_folder1, input_folder2, trans_func, batch_size, train, new_size, return_paths = False, num_workers=4, crop=False, specific_pretransform=None, continuize_transforms=False):
transform = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize(new_size),
transforms.ToTensor(),
transforms.Normalize((0.5,),(0.5,))])
dataset = ImageFolder_Double(input_folder1, input_folder2, transform=transform, return_paths=return_paths, trans_func=trans_func, specific_pretransform=specific_pretransform, continuize_transforms=continuize_transforms)
loader = DataLoader(dataset=dataset, batch_size=batch_size, shuffle=False, drop_last=True, num_workers=num_workers)
return loader
def get_data_loader_folder(input_folder, trans_func, batch_size, train, new_size, return_paths = False, num_workers=4, crop=False):
transform = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize(new_size),
transforms.ToTensor(),
transforms.Normalize((0.5,),(0.5,))])
dataset = ImageFolder(input_folder, transform=transform, return_paths=return_paths, trans_func=trans_func)
loader = DataLoader(dataset=dataset, batch_size=batch_size, shuffle=False, drop_last=True, num_workers=num_workers)
return loader
def get_config(config):
with open(config, 'r') as stream:
return yaml.load(stream)
def eformat(f, prec):
s = "%.*e"%(prec, f)
mantissa, exp = s.split('e')
# add 1 to digits as 1 is taken by sign +/-
return "%se%d"%(mantissa, int(exp))
#def __write_images(image_outputs, display_image_num, file_name):
# image_outputs = [images.expand(-1, 3, -1, -1) for images in image_outputs] # expand gray-scale images to 3 channels
# image_tensor = torch.cat([images[:display_image_num] for images in image_outputs], 0)
# image_grid = vutils.make_grid(image_tensor.data, nrow=display_image_num, padding=0, normalize=True)
# vutils.save_image(image_grid, file_name, nrow=1)
#def write_2images(image_outputs, display_image_num, image_directory, postfix, obj1, obj2):
# n = len(image_outputs)
# __write_images(image_outputs[0:n//2], display_image_num, '%s/gen_%s2%s_%s.jpg' % (image_directory, obj1, obj2, postfix))
# __write_images(image_outputs[n//2:n], display_image_num, '%s/gen_%s2%s_%s.jpg' % (image_directory, obj2, obj1, postfix))
def write_2images(image_outputs, display_image_num, image_directory, postfix, obj):
file_name = '%s/gen_%s_%s.jpg' % (image_directory, obj, postfix)
image_outputs = [images.expand(-1, 3, -1, -1) for images in image_outputs] # expand gray-scale images to 3 channels
image_tensor = torch.cat([images[:display_image_num] for images in image_outputs], 0)
image_grid = vutils.make_grid(image_tensor.data, nrow=display_image_num, padding=0, normalize=True)
vutils.save_image(image_grid, file_name, nrow=1)
def prepare_sub_folder(output_directory):
image_directory = os.path.join(output_directory, 'images')
if not os.path.exists(image_directory):
print("Creating directory: {}".format(image_directory))
os.makedirs(image_directory)
checkpoint_directory = os.path.join(output_directory, 'checkpoints')
if not os.path.exists(checkpoint_directory):
print("Creating directory: {}".format(checkpoint_directory))
os.makedirs(checkpoint_directory)
return checkpoint_directory, image_directory
def write_one_row_html(html_file, iterations, img_filename, all_size):
html_file.write("<h3>iteration [%d] (%s)</h3>" % (iterations,img_filename.split('/')[-1]))
html_file.write("""
<p><a href="%s">
<img src="%s" style="width:%dpx">
</a><br>
<p>
""" % (img_filename, img_filename, all_size))
return
def write_html(filename, iterations, image_save_iterations, image_directory, all_size=1536):
html_file = open(filename, "w")
html_file.write('''
<!DOCTYPE html>
<html>
<head>
<title>Experiment name = %s</title>
<meta http-equiv="refresh" content="30">
</head>
<body>
''' % os.path.basename(filename))
html_file.write("<h3>current</h3>")
write_one_row_html(html_file, iterations, '%s/gen_a_train_current.jpg' % (image_directory), all_size)
write_one_row_html(html_file, iterations, '%s/gen_b_train_current.jpg' % (image_directory), all_size)
for j in range(iterations, image_save_iterations-1, -1):
if j % image_save_iterations == 0:
write_one_row_html(html_file, j, '%s/gen_a_train_%08d.jpg' % (image_directory, j), all_size)
write_one_row_html(html_file, j, '%s/gen_b_train_%08d.jpg' % (image_directory, j), all_size)
write_one_row_html(html_file, j, '%s/gen__test_%08d.jpg' % (image_directory, j), all_size)
html_file.write("</body></html>")
html_file.close()
def write_loss(iterations, trainer, train_writer):
members = [attr for attr in dir(trainer) \
if not callable(getattr(trainer, attr)) and not attr.startswith("__") and ('loss' in attr or 'grad' in attr or 'nwd' in attr)]
for m in members:
train_writer.add_scalar(m, getattr(trainer, m), iterations + 1)
def slerp(val, low, high):
"""
original: Animating Rotation with Quaternion Curves, Ken Shoemake
https://arxiv.org/abs/1609.04468
Code: https://github.com/soumith/dcgan.torch/issues/14, Tom White
"""
omega = np.arccos(np.dot(low / np.linalg.norm(low), high / np.linalg.norm(high)))
so = np.sin(omega)
return np.sin((1.0 - val) * omega) / so * low + np.sin(val * omega) / so * high
def get_slerp_interp(nb_latents, nb_interp, z_dim):
"""
modified from: PyTorch inference for "Progressive Growing of GANs" with CelebA snapshot
https://github.com/ptrblck/prog_gans_pytorch_inference
"""
latent_interps = np.empty(shape=(0, z_dim), dtype=np.float32)
for _ in range(nb_latents):
low = np.random.randn(z_dim)
high = np.random.randn(z_dim) # low + np.random.randn(512) * 0.7
interp_vals = np.linspace(0, 1, num=nb_interp)
latent_interp = np.array([slerp(v, low, high) for v in interp_vals],
dtype=np.float32)
latent_interps = np.vstack((latent_interps, latent_interp))
return latent_interps[:, :, np.newaxis, np.newaxis]
# Get model list for resume
def get_model_list(dirname, key):
if os.path.exists(dirname) is False:
return None
gen_models = [os.path.join(dirname, f) for f in os.listdir(dirname) if
os.path.isfile(os.path.join(dirname, f)) and key in f and ".pt" in f]
if gen_models is None:
return None
gen_models.sort()
last_model_name = gen_models[-1]
return last_model_name
def load_vgg16(model_dir):
""" Use the model from https://github.com/abhiskk/fast-neural-style/blob/master/neural_style/utils.py """
if not os.path.exists(model_dir):
os.mkdir(model_dir)
if not os.path.exists(os.path.join(model_dir, 'vgg16.weight')):
if not os.path.exists(os.path.join(model_dir, 'vgg16.t7')):
os.system('wget https://www.dropbox.com/s/76l3rt4kyi3s8x7/vgg16.t7?dl=1 -O ' + os.path.join(model_dir, 'vgg16.t7'))
vgglua = load_lua(os.path.join(model_dir, 'vgg16.t7'))
vgg = Vgg16()
for (src, dst) in zip(vgglua.parameters()[0], vgg.parameters()):
dst.data[:] = src
torch.save(vgg.state_dict(), os.path.join(model_dir, 'vgg16.weight'))
vgg = Vgg16()
vgg.load_state_dict(torch.load(os.path.join(model_dir, 'vgg16.weight')))
return vgg
def load_inception(model_path):
state_dict = torch.load(model_path)
model = inception_v3(pretrained=False, transform_input=True)
model.aux_logits = False
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, state_dict['fc.weight'].size(0))
model.load_state_dict(state_dict)
for param in model.parameters():
param.requires_grad = False
return model
def vgg_preprocess(batch):
tensortype = type(batch.data)
(r, g, b) = torch.chunk(batch, 3, dim = 1)
batch = torch.cat((b, g, r), dim = 1) # convert RGB to BGR
batch = (batch + 1) * 255 * 0.5 # [-1, 1] -> [0, 255]
mean = tensortype(batch.data.size()).cuda()
mean[:, 0, :, :] = 103.939
mean[:, 1, :, :] = 116.779
mean[:, 2, :, :] = 123.680
batch = batch.sub(Variable(mean)) # subtract mean
return batch
def get_scheduler(optimizer, hyperparameters, iterations=-1):
if 'lr_policy' not in hyperparameters or hyperparameters['lr_policy'] == 'constant':
scheduler = None # constant scheduler
elif hyperparameters['lr_policy'] == 'step':
scheduler = lr_scheduler.StepLR(optimizer, step_size=hyperparameters['step_size'],
gamma=hyperparameters['gamma'], last_epoch=iterations)
else:
return NotImplementedError('learning rate policy [%s] is not implemented', hyperparameters['lr_policy'])
return scheduler
def weights_init(init_type='gaussian'):
def init_fun(m):
classname = m.__class__.__name__
if (classname.find('Conv') == 0 or classname.find('Linear') == 0) and hasattr(m, 'weight'):
# print m.__class__.__name__
if init_type == 'gaussian':
init.normal_(m.weight.data, 0.0, 0.02)
elif init_type == 'xavier':
init.xavier_normal_(m.weight.data, gain=math.sqrt(2))
elif init_type == 'kaiming':
init.kaiming_normal_(m.weight.data, a=0, mode='fan_in')
elif init_type == 'orthogonal':
init.orthogonal_(m.weight.data, gain=math.sqrt(2))
elif init_type == 'default':
pass
else:
assert 0, "Unsupported initialization: {}".format(init_type)
if hasattr(m, 'bias') and m.bias is not None:
init.constant_(m.bias.data, 0.0)
return init_fun
class Timer:
def __init__(self, msg):
self.msg = msg
self.start_time = None
def __enter__(self):
self.start_time = time.time()
def __exit__(self, exc_type, exc_value, exc_tb):
print(self.msg % (time.time() - self.start_time))
def pytorch03_to_pytorch04(state_dict_base, trainer_name):
def __conversion_core(state_dict_base, trainer_name):
state_dict = state_dict_base.copy()
if trainer_name == 'STAR':
for key, value in state_dict_base.items():
if key.endswith(('enc_content.model.0.norm.running_mean',
'enc_content.model.0.norm.running_var',
'enc_content.model.1.norm.running_mean',
'enc_content.model.1.norm.running_var',
'enc_content.model.2.norm.running_mean',
'enc_content.model.2.norm.running_var',
'enc_content.model.3.model.0.model.1.norm.running_mean',
'enc_content.model.3.model.0.model.1.norm.running_var',
'enc_content.model.3.model.0.model.0.norm.running_mean',
'enc_content.model.3.model.0.model.0.norm.running_var',
'enc_content.model.3.model.1.model.1.norm.running_mean',
'enc_content.model.3.model.1.model.1.norm.running_var',
'enc_content.model.3.model.1.model.0.norm.running_mean',
'enc_content.model.3.model.1.model.0.norm.running_var',
'enc_content.model.3.model.2.model.1.norm.running_mean',
'enc_content.model.3.model.2.model.1.norm.running_var',
'enc_content.model.3.model.2.model.0.norm.running_mean',
'enc_content.model.3.model.2.model.0.norm.running_var',
'enc_content.model.3.model.3.model.1.norm.running_mean',
'enc_content.model.3.model.3.model.1.norm.running_var',
'enc_content.model.3.model.3.model.0.norm.running_mean',
'enc_content.model.3.model.3.model.0.norm.running_var',
)):
del state_dict[key]
else:
def __conversion_core(state_dict_base):
state_dict = state_dict_base.copy()
for key, value in state_dict_base.items():
if key.endswith(('enc.model.0.norm.running_mean',
'enc.model.0.norm.running_var',
'enc.model.1.norm.running_mean',
'enc.model.1.norm.running_var',
'enc.model.2.norm.running_mean',
'enc.model.2.norm.running_var',
'enc.model.3.model.0.model.1.norm.running_mean',
'enc.model.3.model.0.model.1.norm.running_var',
'enc.model.3.model.0.model.0.norm.running_mean',
'enc.model.3.model.0.model.0.norm.running_var',
'enc.model.3.model.1.model.1.norm.running_mean',
'enc.model.3.model.1.model.1.norm.running_var',
'enc.model.3.model.1.model.0.norm.running_mean',
'enc.model.3.model.1.model.0.norm.running_var',
'enc.model.3.model.2.model.1.norm.running_mean',
'enc.model.3.model.2.model.1.norm.running_var',
'enc.model.3.model.2.model.0.norm.running_mean',
'enc.model.3.model.2.model.0.norm.running_var',
'enc.model.3.model.3.model.1.norm.running_mean',
'enc.model.3.model.3.model.1.norm.running_var',
'enc.model.3.model.3.model.0.norm.running_mean',
'enc.model.3.model.3.model.0.norm.running_var',
'dec.model.0.model.0.model.1.norm.running_mean',
'dec.model.0.model.0.model.1.norm.running_var',
'dec.model.0.model.0.model.0.norm.running_mean',
'dec.model.0.model.0.model.0.norm.running_var',
'dec.model.0.model.1.model.1.norm.running_mean',
'dec.model.0.model.1.model.1.norm.running_var',
'dec.model.0.model.1.model.0.norm.running_mean',
'dec.model.0.model.1.model.0.norm.running_var',
'dec.model.0.model.2.model.1.norm.running_mean',
'dec.model.0.model.2.model.1.norm.running_var',
'dec.model.0.model.2.model.0.norm.running_mean',
'dec.model.0.model.2.model.0.norm.running_var',
'dec.model.0.model.3.model.1.norm.running_mean',
'dec.model.0.model.3.model.1.norm.running_var',
'dec.model.0.model.3.model.0.norm.running_mean',
'dec.model.0.model.3.model.0.norm.running_var',
)):
del state_dict[key]
return state_dict
state_dict = dict()
state_dict['a'] = __conversion_core(state_dict_base['a'], trainer_name)
state_dict['b'] = __conversion_core(state_dict_base['b'], trainer_name)
return state_dict