-
Notifications
You must be signed in to change notification settings - Fork 14
/
Copy pathconfig_system.py
188 lines (159 loc) · 8.05 KB
/
config_system.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
"""The configuration system."""
import argparse
from fractions import Fraction
import math
import os
from pathlib import Path
import re
import subprocess
import sys
import numpy as np
CONFIG_PY = Path(__file__).parent.resolve() / 'config.py'
def detect_devices():
try:
gpu_list = subprocess.run(['nvidia-smi', '-L'], stdout=subprocess.PIPE,
check=True, universal_newlines=True)
gpus = list(map(int, re.findall(r'^GPU (\d+)', gpu_list.stdout, re.M)))
return gpus if gpus else [-1]
except (subprocess.CalledProcessError, FileNotFoundError):
return [-1]
def ffloat(s):
"""Parses fractional or floating point input strings."""
return float(Fraction(s))
class arg:
def __init__(self, *args, **kwargs):
self.args, self.kwargs = args, kwargs
def add_args(parser, args):
for a in args:
parser.add_argument(*a.args, **a.kwargs)
def parse_args(state_obj=None):
"""Parses command line arguments."""
parser = argparse.ArgumentParser(description='Neural style transfer using Caffe.',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
add_args(parser, [
arg('--content-image', '-ci', help='the content image'),
arg('--style-images', '-si', nargs='+', default=[], metavar='STYLE_IMAGE',
help='the style images'),
arg('--output-image', '-oi', help='the output image'),
arg('--init-image', '-ii', metavar='IMAGE', help='the initial image'),
arg('--aux-image', '-ai', metavar='IMAGE', help='the auxiliary image'),
arg('--config', type=Path, help='a Python source file containing configuration options'),
arg('--list-layers', action='store_true', help='list the model\'s layers'),
arg('--caffe-path', help='the path to the Caffe installation'),
arg('--devices', nargs='+', metavar='DEVICE', type=int, default=[-1],
help='GPU device numbers to use (-1 for cpu)'),
arg('--iterations', '-i', nargs='+', type=int, default=[200, 100],
help='the number of iterations'),
arg('--size', '-s', type=int, default=256, help='the output size'),
arg('--min-size', type=int, default=182, help='the minimum scale\'s size'),
arg('--style-scale', '-ss', type=ffloat, default=1, help='the style scale factor'),
arg('--max-style-size', type=int, help='the maximum style size'),
arg('--style-scale-up', default=False, action='store_true',
help='allow scaling style images up'),
arg('--style-multiscale', '-sm', type=int, nargs=2, metavar=('MIN_SCALE', 'MAX_SCALE'),
default=None, help='combine styles computed at all scales into a single style'),
arg('--tile-size', type=int, default=512, help='the maximum rendering tile size'),
arg('--optimizer', '-o', default='adam', choices=['adam', 'lbfgs'],
help='the optimizer to use'),
arg('--step-size', '-st', type=ffloat, default=15,
help='the initial step size for Adam'),
arg('--step-decay', '-sd', nargs=2, metavar=('DECAY', 'POWER'), type=ffloat,
default=[0.05, 0.5], help='on step i, divide step_size by (1 + DECAY * i)^POWER'),
arg('--avg-window', type=ffloat, default=20, help='the iterate averaging window size'),
arg('--layer-weights', help='a json file containing per-layer weight scaling factors'),
arg('--content-weight', '-cw', type=ffloat, default=0.05, help='the content image factor'),
arg('--dd-weight', '-dw', type=ffloat, default=0, help='the Deep Dream factor'),
arg('--tv-weight', '-tw', type=ffloat, default=5, help='the TV smoothing factor'),
arg('--tv-power', '-tp', metavar='BETA', type=ffloat, default=2,
help='the TV smoothing exponent'),
arg('--swt-weight', '-ww', metavar='WEIGHT', type=ffloat, default=0,
help='the SWT smoothing factor'),
arg('--swt-wavelet', '-wt', metavar='WAVELET', default='haar',
help='the SWT wavelet'),
arg('--swt-levels', '-wl', metavar='LEVELS', default=1, type=int,
help='the number of levels to use for decomposition'),
arg('--swt-power', '-wp', metavar='P', default=2, type=ffloat,
help='the SWT smoothing exponent'),
arg('--p-weight', '-pw', type=ffloat, default=2, help='the p-norm regularizer factor'),
arg('--p-power', '-pp', metavar='P', type=ffloat, default=6, help='the p-norm exponent'),
arg('--aux-weight', '-aw', type=ffloat, default=10, help='the auxiliary image factor'),
arg('--content-layers', nargs='*', default=['conv4_2'], metavar='LAYER',
help='the layers to use for content'),
arg('--style-layers', nargs='*', metavar='LAYER',
default=['conv1_1', 'conv2_1', 'conv3_1', 'conv4_1', 'conv5_1'],
help='the layers to use for style'),
arg('--dd-layers', nargs='*', metavar='LAYER', default=[],
help='the layers to use for Deep Dream'),
arg('--port', '-p', type=int, default=8000, help='the port to use for the http server'),
arg('--display', default='browser', choices=['browser', 'gui', 'none'],
help='the display method to use'),
arg('--browser', default=None, help='the web browser to open the web interface in'),
arg('--model', default='vgg19.prototxt',
help='the Caffe deploy.prototxt for the model to use'),
arg('--weights', default='vgg19.caffemodel',
help='the Caffe .caffemodel for the model to use'),
arg('--mean', nargs=3, metavar=('B_MEAN', 'G_MEAN', 'R_MEAN'),
default=(103.939, 116.779, 123.68),
help='the per-channel means of the model (BGR order)'),
arg('--save-every', metavar='N', type=int, default=0, help='save the image every n steps'),
arg('--seed', type=int, default=0, help='the random seed'),
arg('--div', metavar='FACTOR', type=int, default=1,
help='Ensure all images are divisible by FACTOR '
'(can fix some GPU memory alignment issues)'),
arg('--jitter', action='store_true',
help='use slower but higher quality translation-invariant rendering'),
arg('--debug', action='store_true', help='enable debug messages'),
])
defaults = vars(parser.parse_args([]))
config_args = {}
if CONFIG_PY.exists():
config_args.update(eval_config(CONFIG_PY))
sysv_args = vars(parser.parse_args())
config2_args = {}
if sysv_args['config']:
config2_args.update(eval_config(sysv_args['config']))
args = {}
args.update(defaults)
args.update(config_args)
for a, value in sysv_args.items():
if defaults[a] != value:
args[a] = value
args.update(config2_args)
args2 = AutocallNamespace(state_obj, **args)
if args2.debug:
os.environ['DEBUG'] = '1'
if not args2.list_layers and (not args2.content_image or not args2.style_images):
parser.print_help()
sys.exit(1)
return args2
class ValuePlaceholder:
pass
class AutocallNamespace:
def __init__(self, state_obj, **kwargs):
self.state_obj = state_obj
self.ns = argparse.Namespace(**kwargs)
def __getattr__(self, name):
value = getattr(self.ns, name)
if callable(value):
try:
return value(self.state_obj)
except AttributeError:
return ValuePlaceholder()
return value
def __setattr__(self, name, value):
if name in ('state_obj', 'ns'):
object.__setattr__(self, name, value)
return
setattr(self.ns, name, value)
def __iter__(self):
yield from vars(self.ns)
def __contains__(self, key):
return key in self.ns
def __repr__(self):
return 'Autocall' + repr(self.ns)
CONFIG_GLOBALS = dict(detect_devices=detect_devices, math=math, np=np)
def eval_config(config_file):
config_code = compile(config_file.read_text(), config_file.name, 'exec')
locs = {}
exec(config_code, CONFIG_GLOBALS, locs) # pylint: disable=exec-used
return locs