-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
executable file
·564 lines (472 loc) · 25 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
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
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
# Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
#
# NVIDIA CORPORATION and its licensors retain all intellectual property
# and proprietary rights in and to this software, related documentation
# and any modifications thereto. Any use, reproduction, disclosure or
# distribution of this software and related documentation without an express
# license agreement from NVIDIA CORPORATION is strictly prohibited.
"""Train a GAN using the techniques described in the paper
"Training Generative Adversarial Networks with Limited Data"."""
import os
import click
import re
import json
import tempfile
import torch
import dnnlib
from training import training_loop
from metrics import metric_main
from torch_utils import training_stats
from torch_utils import custom_ops
ENCODERS = ['RN50', 'RN101', 'RN50x4', 'RN50x16', 'RN50x64',
'ViT-B/32', 'ViT-B/16', 'ViT-L/14', 'ViT-L/14@336px']
encoder_dims = {model:512 if 'ViT-L' not in model else 768 for model in ENCODERS}
#----------------------------------------------------------------------------
class UserError(Exception):
pass
#----------------------------------------------------------------------------
def setup_training_loop_kwargs(
# General options (not included in desc).
gpus = None, # Number of GPUs: <int>, default = 1 gpu
snap = None, # Snapshot interval: <int>, default = 50 ticks
metrics = None, # List of metric names: [], ['fid50k_full'] (default), ...
seed = None, # Random seed: <int>, default = 0
# Dataset.
data = None, # Training dataset (required): <path>
condition = None, # Conditional Image path
cond = None, # Train conditional model based on dataset labels: <bool>, default = False
subset = None, # Train with only N images: <int>, default = all
mirror = None, # Augment dataset with x-flips: <bool>, default = False
# Image encoder
encoder = None, # Image encoder name: <str>, default = 'ViT-B/32'. Valid when cond is not None.
# Base config.
cfg = None, # Base config: 'auto' (default), 'stylegan2', 'paper256', 'paper512', 'paper1024', 'cifar'
gamma = None, # Override R1 gamma: <float>
kimg = None, # Override training duration: <int>
batch = None, # Override batch size: <int>
# Discriminator augmentation.
aug = None, # Augmentation mode: 'ada' (default), 'noaug', 'fixed'
p = None, # Specify p for 'fixed' (required): <float>
target = None, # Override ADA target for 'ada': <float>, default = depends on aug
augpipe = None, # Augmentation pipeline: 'blit', 'geom', 'color', 'filter', 'noise', 'cutout', 'bg', 'bgc' (default), ..., 'bgcfnc'
# Transfer learning.
resume = None, # Load previous network: 'noresume' (default), 'ffhq256', 'ffhq512', 'ffhq1024', 'celebahq256', 'lsundog256', <file>, <url>
freezed = None, # Freeze-D: <int>, default = 0 discriminator layers
# Performance options (not included in desc).
fp32 = None, # Disable mixed-precision training: <bool>, default = False
nhwc = None, # Use NHWC memory format with FP16: <bool>, default = False
allow_tf32 = None, # Allow PyTorch to use TF32 for matmul and convolutions: <bool>, default = False
nobench = None, # Disable cuDNN benchmarking: <bool>, default = False
workers = None, # Override number of DataLoader workers: <int>, default = 3
):
args = dnnlib.EasyDict()
# ------------------------------------------
# General options: gpus, snap, metrics, seed
# ------------------------------------------
if gpus is None:
gpus = 1
assert isinstance(gpus, int)
if not (gpus >= 1 and gpus & (gpus - 1) == 0):
raise UserError('--gpus must be a power of two')
args.num_gpus = gpus
if snap is None:
snap = 50
assert isinstance(snap, int)
if snap < 1:
raise UserError('--snap must be at least 1')
args.image_snapshot_ticks = snap
args.network_snapshot_ticks = snap
if metrics is None:
metrics = ['fid50k_full']
assert isinstance(metrics, list)
if not all(metric_main.is_valid_metric(metric) for metric in metrics):
raise UserError('\n'.join(['--metrics can only contain the following values:'] + metric_main.list_valid_metrics()))
args.metrics = metrics
if seed is None:
seed = 0
assert isinstance(seed, int)
args.random_seed = seed
# -----------------------------------
# Dataset: data, cond, subset, mirror
# -----------------------------------
assert data is not None
assert isinstance(data, str)
if condition == None:
args.training_set_kwargs = dnnlib.EasyDict(class_name='training.dataset.ImageFolderDataset', path=data, use_labels=True, max_size=None, xflip=False)
img_dim = 0
args.image_encoder_kwargs = dnnlib.EasyDict()
else:
args.training_set_kwargs = dnnlib.EasyDict(class_name='training.dataset.ImageConditionalDataset', pair_path=data, condition_path=condition, use_labels=True, max_size=None, xflip=False)
if encoder is None:
encoder = 'ViT-B/32'
assert encoder in ENCODERS
img_dim = encoder_dims[encoder]
args.image_encoder_kwargs = dnnlib.EasyDict(name=encoder)
args.data_loader_kwargs = dnnlib.EasyDict(pin_memory=True, num_workers=3, prefetch_factor=3)
try:
training_set = dnnlib.util.construct_class_by_name(**args.training_set_kwargs) # subclass of training.dataset.Dataset
args.training_set_kwargs.resolution = training_set.resolution # be explicit about resolution
args.training_set_kwargs.use_labels = training_set.has_labels # be explicit about labels
args.training_set_kwargs.max_size = len(training_set) # be explicit about dataset size
desc = training_set.name
del training_set # conserve memory
except IOError as err:
raise UserError(f'--data: {err}')
if cond is None:
cond = False
assert isinstance(cond, bool)
if cond:
if not args.training_set_kwargs.use_labels:
raise UserError('--cond=True requires labels specified in dataset.json')
desc += '-cond'
else:
args.training_set_kwargs.use_labels = False
if subset is not None:
assert isinstance(subset, int)
if not 1 <= subset <= args.training_set_kwargs.max_size:
raise UserError(f'--subset must be between 1 and {args.training_set_kwargs.max_size}')
desc += f'-subset{subset}'
if subset < args.training_set_kwargs.max_size:
args.training_set_kwargs.max_size = subset
args.training_set_kwargs.random_seed = args.random_seed
if mirror is None:
mirror = False
assert isinstance(mirror, bool)
if mirror:
desc += '-mirror'
args.training_set_kwargs.xflip = True
# ------------------------------------
# Base config: cfg, gamma, kimg, batch
# ------------------------------------
if cfg is None:
cfg = 'auto'
assert isinstance(cfg, str)
desc += f'-{cfg}'
cfg_specs = {
'auto': dict(ref_gpus=-1, kimg=25000, mb=-1, mbstd=-1, fmaps=-1, lrate=-1, gamma=-1, ema=-1, ramp=0.05, map=2), # Populated dynamically based on resolution and GPU count.
'stylegan2': dict(ref_gpus=8, kimg=25000, mb=32, mbstd=4, fmaps=1, lrate=0.002, gamma=10, ema=10, ramp=None, map=8), # Uses mixed-precision, unlike the original StyleGAN2.
'paper256': dict(ref_gpus=8, kimg=25000, mb=64, mbstd=8, fmaps=0.5, lrate=0.0025, gamma=1, ema=20, ramp=None, map=8),
'paper512': dict(ref_gpus=8, kimg=25000, mb=64, mbstd=8, fmaps=1, lrate=0.0025, gamma=0.5, ema=20, ramp=None, map=8),
'paper1024': dict(ref_gpus=8, kimg=25000, mb=32, mbstd=4, fmaps=1, lrate=0.002, gamma=2, ema=10, ramp=None, map=8),
'cifar': dict(ref_gpus=2, kimg=100000, mb=64, mbstd=32, fmaps=1, lrate=0.0025, gamma=0.01, ema=500, ramp=0.05, map=2),
}
assert cfg in cfg_specs
spec = dnnlib.EasyDict(cfg_specs[cfg])
if cfg == 'auto':
desc += f'{gpus:d}'
spec.ref_gpus = gpus
res = args.training_set_kwargs.resolution
spec.mb = max(min(gpus * min(4096 // res, 32), 64), gpus) # keep gpu memory consumption at bay
spec.mbstd = min(spec.mb // gpus, 4) # other hyperparams behave more predictably if mbstd group size remains fixed
spec.fmaps = 1 if res >= 512 else 0.5
spec.lrate = 0.002 if res >= 1024 else 0.0025
spec.gamma = 0.0002 * (res ** 2) / spec.mb # heuristic formula
spec.ema = spec.mb * 10 / 32
args.G_kwargs = dnnlib.EasyDict(class_name='training.networks.Generator', z_dim=512, w_dim=512, mapping_kwargs=dnnlib.EasyDict(img_dim=img_dim), synthesis_kwargs=dnnlib.EasyDict())
args.D_kwargs = dnnlib.EasyDict(class_name='training.networks.Discriminator', block_kwargs=dnnlib.EasyDict(), mapping_kwargs=dnnlib.EasyDict(img_dim=img_dim), epilogue_kwargs=dnnlib.EasyDict())
args.G_kwargs.synthesis_kwargs.channel_base = args.D_kwargs.channel_base = int(spec.fmaps * 32768)
args.G_kwargs.synthesis_kwargs.channel_max = args.D_kwargs.channel_max = 512
args.G_kwargs.mapping_kwargs.num_layers = spec.map
args.G_kwargs.synthesis_kwargs.num_fp16_res = args.D_kwargs.num_fp16_res = 4 # enable mixed-precision training
args.G_kwargs.synthesis_kwargs.conv_clamp = args.D_kwargs.conv_clamp = 256 # clamp activations to avoid float16 overflow
args.D_kwargs.epilogue_kwargs.mbstd_group_size = spec.mbstd
args.G_opt_kwargs = dnnlib.EasyDict(class_name='torch.optim.Adam', lr=spec.lrate, betas=[0,0.99], eps=1e-8)
args.D_opt_kwargs = dnnlib.EasyDict(class_name='torch.optim.Adam', lr=spec.lrate, betas=[0,0.99], eps=1e-8)
args.loss_kwargs = dnnlib.EasyDict(class_name='training.loss.StyleGAN2Loss', r1_gamma=spec.gamma, image_encoder_kwargs=dnnlib.EasyDict(name=encoder))
args.total_kimg = spec.kimg
args.batch_size = spec.mb
args.batch_gpu = spec.mb // spec.ref_gpus
args.ema_kimg = spec.ema
args.ema_rampup = spec.ramp
if cfg == 'cifar':
args.loss_kwargs.pl_weight = 0 # disable path length regularization
args.loss_kwargs.style_mixing_prob = 0 # disable style mixing
args.D_kwargs.architecture = 'orig' # disable residual skip connections
if gamma is not None:
assert isinstance(gamma, float)
if not gamma >= 0:
raise UserError('--gamma must be non-negative')
desc += f'-gamma{gamma:g}'
args.loss_kwargs.r1_gamma = gamma
if kimg is not None:
assert isinstance(kimg, int)
if not kimg >= 1:
raise UserError('--kimg must be at least 1')
desc += f'-kimg{kimg:d}'
args.total_kimg = kimg
if batch is not None:
assert isinstance(batch, int)
if not (batch >= 1 and batch % gpus == 0):
raise UserError('--batch must be at least 1 and divisible by --gpus')
desc += f'-batch{batch}'
args.batch_size = batch
args.batch_gpu = batch // gpus
# ---------------------------------------------------
# Discriminator augmentation: aug, p, target, augpipe
# ---------------------------------------------------
if aug is None:
aug = 'ada'
else:
assert isinstance(aug, str)
desc += f'-{aug}'
if aug == 'ada':
args.ada_target = 0.6
elif aug == 'noaug':
pass
elif aug == 'fixed':
if p is None:
raise UserError(f'--aug={aug} requires specifying --p')
else:
raise UserError(f'--aug={aug} not supported')
if p is not None:
assert isinstance(p, float)
if aug != 'fixed':
raise UserError('--p can only be specified with --aug=fixed')
if not 0 <= p <= 1:
raise UserError('--p must be between 0 and 1')
desc += f'-p{p:g}'
args.augment_p = p
if target is not None:
assert isinstance(target, float)
if aug != 'ada':
raise UserError('--target can only be specified with --aug=ada')
if not 0 <= target <= 1:
raise UserError('--target must be between 0 and 1')
desc += f'-target{target:g}'
args.ada_target = target
assert augpipe is None or isinstance(augpipe, str)
if augpipe is None:
augpipe = 'bgc'
else:
if aug == 'noaug':
raise UserError('--augpipe cannot be specified with --aug=noaug')
desc += f'-{augpipe}'
augpipe_specs = {
'blit': dict(xflip=1, rotate90=1, xint=1),
'geom': dict(scale=1, rotate=1, aniso=1, xfrac=1),
'color': dict(brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1),
'filter': dict(imgfilter=1),
'noise': dict(noise=1),
'imgcond': dict(xint=1, scale=1, aniso=1, xfrac=1, brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1, rotate_max=1/12),
'cutout': dict(cutout=1),
'cn': dict(brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1, noise=1),
'bg': dict(xflip=1, rotate90=1, xint=1, scale=1, rotate=1, aniso=1, xfrac=1),
'bgc': dict(xflip=1, rotate90=1, xint=1, scale=1, rotate=1, aniso=1, xfrac=1, brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1),
'bgcf': dict(xflip=1, rotate90=1, xint=1, scale=1, rotate=1, aniso=1, xfrac=1, brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1, imgfilter=1),
'bgcfn': dict(xflip=1, rotate90=1, xint=1, scale=1, rotate=1, aniso=1, xfrac=1, brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1, imgfilter=1, noise=1),
'bgcfnc': dict(xflip=1, rotate90=1, xint=1, scale=1, rotate=1, aniso=1, xfrac=1, brightness=1, contrast=1, lumaflip=1, hue=1, saturation=1, imgfilter=1, noise=1, cutout=1),
}
assert augpipe in augpipe_specs
if aug != 'noaug':
args.augment_kwargs = dnnlib.EasyDict(class_name='training.augment.AugmentPipe', **augpipe_specs[augpipe])
args.cond_augment_kwargs = dnnlib.EasyDict(class_name='training.augment.AugmentPipe', **augpipe_specs['imgcond'])
# ----------------------------------
# Transfer learning: resume, freezed
# ----------------------------------
resume_specs = {
'ffhq256': 'https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/transfer-learning-source-nets/ffhq-res256-mirror-paper256-noaug.pkl',
'ffhq512': 'https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/transfer-learning-source-nets/ffhq-res512-mirror-stylegan2-noaug.pkl',
'ffhq1024': 'https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/transfer-learning-source-nets/ffhq-res1024-mirror-stylegan2-noaug.pkl',
'celebahq256': 'https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/transfer-learning-source-nets/celebahq-res256-mirror-paper256-kimg100000-ada-target0.5.pkl',
'lsundog256': 'https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/transfer-learning-source-nets/lsundog-res256-paper256-kimg100000-noaug.pkl',
}
assert resume is None or isinstance(resume, str)
if resume is None:
resume = 'noresume'
elif resume == 'noresume':
desc += '-noresume'
elif resume in resume_specs:
desc += f'-resume{resume}'
args.resume_pkl = resume_specs[resume] # predefined url
else:
desc += '-resumecustom'
args.resume_pkl = resume # custom path or url
if resume != 'noresume':
args.ada_kimg = 100 # make ADA react faster at the beginning
args.ema_rampup = None # disable EMA rampup
if freezed is not None:
assert isinstance(freezed, int)
if not freezed >= 0:
raise UserError('--freezed must be non-negative')
desc += f'-freezed{freezed:d}'
args.D_kwargs.block_kwargs.freeze_layers = freezed
# -------------------------------------------------
# Performance options: fp32, nhwc, nobench, workers
# -------------------------------------------------
if fp32 is None:
fp32 = False
assert isinstance(fp32, bool)
if fp32:
args.G_kwargs.synthesis_kwargs.num_fp16_res = args.D_kwargs.num_fp16_res = 0
args.G_kwargs.synthesis_kwargs.conv_clamp = args.D_kwargs.conv_clamp = None
if nhwc is None:
nhwc = False
assert isinstance(nhwc, bool)
if nhwc:
args.G_kwargs.synthesis_kwargs.fp16_channels_last = args.D_kwargs.block_kwargs.fp16_channels_last = True
if nobench is None:
nobench = False
assert isinstance(nobench, bool)
if nobench:
args.cudnn_benchmark = False
if allow_tf32 is None:
allow_tf32 = False
assert isinstance(allow_tf32, bool)
if allow_tf32:
args.allow_tf32 = True
if workers is not None:
assert isinstance(workers, int)
if not workers >= 1:
raise UserError('--workers must be at least 1')
args.data_loader_kwargs.num_workers = workers
return desc, args
#----------------------------------------------------------------------------
def subprocess_fn(rank, args, temp_dir):
dnnlib.util.Logger(file_name=os.path.join(args.run_dir, 'log.txt'), file_mode='a', should_flush=True)
# Init torch.distributed.
if args.num_gpus > 1:
init_file = os.path.abspath(os.path.join(temp_dir, '.torch_distributed_init'))
if os.name == 'nt':
init_method = 'file:///' + init_file.replace('\\', '/')
torch.distributed.init_process_group(backend='gloo', init_method=init_method, rank=rank, world_size=args.num_gpus)
else:
init_method = f'file://{init_file}'
torch.distributed.init_process_group(backend='nccl', init_method=init_method, rank=rank, world_size=args.num_gpus)
# Init torch_utils.
sync_device = torch.device('cuda', rank) if args.num_gpus > 1 else None
training_stats.init_multiprocessing(rank=rank, sync_device=sync_device)
if rank != 0:
custom_ops.verbosity = 'none'
# Execute training loop.
training_loop.training_loop(rank=rank, **args)
#----------------------------------------------------------------------------
class CommaSeparatedList(click.ParamType):
name = 'list'
def convert(self, value, param, ctx):
_ = param, ctx
if value is None or value.lower() == 'none' or value == '':
return []
return value.split(',')
#----------------------------------------------------------------------------
@click.command()
@click.pass_context
# General options.
@click.option('--outdir', help='Where to save the results', required=True, metavar='DIR')
@click.option('--gpus', help='Number of GPUs to use [default: 1]', type=int, metavar='INT')
@click.option('--snap', help='Snapshot interval [default: 50 ticks]', type=int, metavar='INT')
@click.option('--metrics', help='Comma-separated list or "none" [default: fid50k_full]', type=CommaSeparatedList())
@click.option('--seed', help='Random seed [default: 0]', type=int, metavar='INT')
@click.option('-n', '--dry-run', help='Print training options and exit', is_flag=True)
# Dataset.
@click.option('--data', help='Training data (directory or zip)', metavar='PATH', required=True)
@click.option('--cond', help='Train conditional model based on dataset labels [default: false]', type=bool, metavar='BOOL')
@click.option('--condition', help='Conditional image path', metavar='PATH', required=True)
@click.option('--subset', help='Train with only N images [default: all]', type=int, metavar='INT')
@click.option('--mirror', help='Enable dataset x-flips [default: false]', type=bool, metavar='BOOL')
# Base config.
@click.option('--cfg', help='Base config [default: auto]', type=click.Choice(['auto', 'stylegan2', 'paper256', 'paper512', 'paper1024', 'cifar']))
@click.option('--gamma', help='Override R1 gamma', type=float)
@click.option('--kimg', help='Override training duration', type=int, metavar='INT')
@click.option('--batch', help='Override batch size', type=int, metavar='INT')
@click.option('--encoder', help='CLIP encoder name', type=click.Choice(ENCODERS))
# Discriminator augmentation.
@click.option('--aug', help='Augmentation mode [default: ada]', type=click.Choice(['noaug', 'ada', 'fixed']))
@click.option('--p', help='Augmentation probability for --aug=fixed', type=float)
@click.option('--target', help='ADA target value for --aug=ada', type=float)
@click.option('--augpipe', help='Augmentation pipeline [default: bgc]', type=click.Choice(['blit', 'geom', 'color', 'filter', 'noise', 'cutout', 'bg', 'bgc', 'bgcf', 'bgcfn', 'bgcfnc']))
# Transfer learning.
@click.option('--resume', help='Resume training [default: noresume]', metavar='PKL')
@click.option('--freezed', help='Freeze-D [default: 0 layers]', type=int, metavar='INT')
# Performance options.
@click.option('--fp32', help='Disable mixed-precision training', type=bool, metavar='BOOL')
@click.option('--nhwc', help='Use NHWC memory format with FP16', type=bool, metavar='BOOL')
@click.option('--nobench', help='Disable cuDNN benchmarking', type=bool, metavar='BOOL')
@click.option('--allow-tf32', help='Allow PyTorch to use TF32 internally', type=bool, metavar='BOOL')
@click.option('--workers', help='Override number of DataLoader workers', type=int, metavar='INT')
def main(ctx, outdir, dry_run, **config_kwargs):
"""Train a GAN using the techniques described in the paper
"Training Generative Adversarial Networks with Limited Data".
Examples:
\b
# Train with custom dataset using 1 GPU.
python train.py --outdir=~/training-runs --data=~/mydataset.zip --gpus=1
\b
# Train class-conditional CIFAR-10 using 2 GPUs.
python train.py --outdir=~/training-runs --data=~/datasets/cifar10.zip \\
--gpus=2 --cfg=cifar --cond=1
\b
# Transfer learn MetFaces from FFHQ using 4 GPUs.
python train.py --outdir=~/training-runs --data=~/datasets/metfaces.zip \\
--gpus=4 --cfg=paper1024 --mirror=1 --resume=ffhq1024 --snap=10
\b
# Reproduce original StyleGAN2 config F.
python train.py --outdir=~/training-runs --data=~/datasets/ffhq.zip \\
--gpus=8 --cfg=stylegan2 --mirror=1 --aug=noaug
\b
Base configs (--cfg):
auto Automatically select reasonable defaults based on resolution
and GPU count. Good starting point for new datasets.
stylegan2 Reproduce results for StyleGAN2 config F at 1024x1024.
paper256 Reproduce results for FFHQ and LSUN Cat at 256x256.
paper512 Reproduce results for BreCaHAD and AFHQ at 512x512.
paper1024 Reproduce results for MetFaces at 1024x1024.
cifar Reproduce results for CIFAR-10 at 32x32.
\b
Transfer learning source networks (--resume):
ffhq256 FFHQ trained at 256x256 resolution.
ffhq512 FFHQ trained at 512x512 resolution.
ffhq1024 FFHQ trained at 1024x1024 resolution.
celebahq256 CelebA-HQ trained at 256x256 resolution.
lsundog256 LSUN Dog trained at 256x256 resolution.
<PATH or URL> Custom network pickle.
"""
dnnlib.util.Logger(should_flush=True)
# Setup training options.
try:
run_desc, args = setup_training_loop_kwargs(**config_kwargs)
except UserError as err:
ctx.fail(err)
# Pick output directory.
prev_run_dirs = []
if os.path.isdir(outdir):
prev_run_dirs = [x for x in os.listdir(outdir) if os.path.isdir(os.path.join(outdir, x))]
prev_run_ids = [re.match(r'^\d+', x) for x in prev_run_dirs]
prev_run_ids = [int(x.group()) for x in prev_run_ids if x is not None]
cur_run_id = max(prev_run_ids, default=-1) + 1
args.run_dir = os.path.join(outdir, f'{cur_run_id:05d}-{run_desc}')
assert not os.path.exists(args.run_dir)
# Print options.
print()
print('Training options:')
print(json.dumps(args, indent=2))
print()
print(f'Output directory: {args.run_dir}')
print(f'Training data: {args.training_set_kwargs.pair_path}')
print(f'Training duration: {args.total_kimg} kimg')
print(f'Number of GPUs: {args.num_gpus}')
print(f'Number of images: {args.training_set_kwargs.max_size}')
print(f'Image resolution: {args.training_set_kwargs.resolution}')
print(f'Conditional model: {args.training_set_kwargs.use_labels}')
print(f'Dataset x-flips: {args.training_set_kwargs.xflip}')
print()
# Dry run?
if dry_run:
print('Dry run; exiting.')
return
# Create output directory.
print('Creating output directory...')
os.makedirs(args.run_dir)
with open(os.path.join(args.run_dir, 'training_options.json'), 'wt') as f:
json.dump(args, f, indent=2)
# Launch processes.
print('Launching processes...')
torch.multiprocessing.set_start_method('spawn')
with tempfile.TemporaryDirectory() as temp_dir:
if args.num_gpus == 1:
subprocess_fn(rank=0, args=args, temp_dir=temp_dir)
else:
torch.multiprocessing.spawn(fn=subprocess_fn, args=(args, temp_dir), nprocs=args.num_gpus)
#----------------------------------------------------------------------------
if __name__ == "__main__":
main() # pylint: disable=no-value-for-parameter
#----------------------------------------------------------------------------