-
Notifications
You must be signed in to change notification settings - Fork 13
/
parameter.py
173 lines (161 loc) · 9.58 KB
/
parameter.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
from argparse import ArgumentParser, Namespace
def parse_arguments() -> Namespace:
parser = ArgumentParser()
# dataset
parser.add_argument('--imgdir', type=str, required=True, default='./datasets/',
help='Directory containing the processed data')
parser.add_argument('--outdir', type=str, required=False,
help='Subfolder in ./results/ for saving.')
parser.add_argument('--imgname', type=str, help='The name of original images')
parser.add_argument('--maskname', type=str, help='The name of corrupted images')
parser.add_argument('--gain', type=float, required=False, default=2e3,
help='gain for the input')
parser.add_argument('--datadim', type=str, required=False, default='2d', choices=['2d', '2.5d', '3d'],
help='The dimensionality of the data')
parser.add_argument('--slice', type=str, required=False, default='xy', choices=['tx', 'ty', 'xy'],
help='The type of slice of 3D data when datadim=2.5d')
parser.add_argument('--imgchannel', type=int, required=False,
help='Number of 2.5d patches to be stacked in the channel dimension.')
parser.add_argument('--adirandel', type=float, required=False, default=0.,
help='The percent of addictive random deleting samples')
parser.add_argument('--padwidth', type=int, required=False, default=0,
help='The padding width to the process data using edge mode')
parser.add_argument('--patch_shape', nargs='+', type=int, required=False,
help="Patch shape to be processed (it can handle 2D, 2.5D, 3D)")
parser.add_argument('--patch_stride', nargs='+', type=int, required=False,
help="Patch stride for the extraction (it can handle 2D, 2.5D, 3D)")
# network design
parser.add_argument('--net', type=str, required=False, default='multiunet',
choices=['multiunet', 'attmultiunet', 'part', 'unet', 'load'],
help='The network architecture')
parser.add_argument('--gpu', type=int, required=False,
help='GPU to use (default lowest memory usage)')
parser.add_argument('--activation', type=str, default='LeakyReLU', required=False,
choices=['LeakyReLU', 'ReLU', 'ELU', 'Tanh', 'Sigmoid'],
help='Activation function to be used in the convolution block')
parser.add_argument('--last_activation', type=str, required=False,
choices=['LeakyReLU', 'ReLU', 'ELU', 'Tanh', 'Sigmoid'],
help='Activation function to the network output')
parser.add_argument('--dropout', type=float, default=0., required=False,
help='Dropout rate to be applied in each convolution')
parser.add_argument('--filters', nargs='+', type=int, required=False, default=[16, 32, 64, 128, 256],
help='Numbers of channels in every layer of encoder and decoder')
parser.add_argument('--skip', nargs='+', type=int, required=False, default=[16, 32, 64, 128],
help='Number of channels for skip-connection')
parser.add_argument('--inputdepth', type=int, required=False, default=64,
help='Depth of the input noise tensor')
parser.add_argument('--upsample', type=str, required=False, default='nearest',
choices=['nearest', 'linear'],
help="Network's upgoing deconvolution strategy")
parser.add_argument('--inittype', type=str, required=False, default='xavier',
choices=['xavier', 'normal', 'default', 'kaiming', 'orthogonal'],
help='Initialization strategy for the network weights')
parser.add_argument('--initgain', type=float, required=False, default=0.02,
help='Initialization scaling factor for normal, xavier and orthogonal.')
parser.add_argument('--savemodel', action='store_true', default=False,
help='Save the optimized model to disk')
parser.add_argument('--netdir', type=str, nargs='+', required=False,
help='Path for loading the optimized network')
# input noise
parser.add_argument('--param_noise', action='store_false',
help='Add normal noise to the parameters every epoch')
parser.add_argument('--reg_noise_std', type=float, required=False, default=0.03,
help='Standard deviation of the normal noise to be added to the input every epoch')
parser.add_argument('--noise_dist', type=str, default='n', required=False, choices=['n', 'u', 'c'],
help='Type of noise for the input tensor [(n)ormal, (u)niform, (c)auchy]')
parser.add_argument('--noise_std', type=float, default=.1, required=False,
help='Standard deviation of the noise for the input tensor')
parser.add_argument('--data_forgetting_factor', type=int, default=0, required=False,
help='Duration of additional decimated data to the input noise tensor')
parser.add_argument('--filter_noise_with_wavelet', action='store_true', default=False,
help='Filter input noise tensor with the wavelet bandwidth')
parser.add_argument('--lowpass_fs', type=float, required=False,
help='Filter input noise tensor with a 4th order Butterworth LPF: sampling frequency')
parser.add_argument('--lowpass_fc', type=float, required=False,
help='Filter input noise tensor with a 4th order Butterworth LPF: cutoff frequency')
parser.add_argument('--lowpass_ntaps', type=int, required=False, default=7,
help='Low pass filter lenght')
# training
parser.add_argument('--loss', type=str, required=False, choices=['mae', 'mse'], default='mae',
help='Loss function to be used.')
parser.add_argument('--epochs', '-e', '--iter', type=int, required=False, default=2001,
help='Number of optimization iterations')
parser.add_argument('--lr', type=float, default=1e-3, required=False,
help='Learning Rate for Adam optimizer')
parser.add_argument('--lr_factor', type=float, default=.9, required=False,
help='LR reduction for Plateau scheduler.')
parser.add_argument('--lr_thresh', type=float, default=1e-5, required=False,
help='LR threshold for Plateau scheduler.')
parser.add_argument('--lr_patience', type=int, default=100, required=False,
help='LR patience for Plateau scheduler.')
parser.add_argument('--save_every', type=int, required=False,
help='Number of epochs every which to save the results')
parser.add_argument('--start_from_prev', action='store_true', default=False,
help='Start training from previous patch')
parser.add_argument('--reduce_lr', action='store_true', default=False,
help='Use ReduceLROnPlateau scheduler')
parser.add_argument('--earlystop_patience', type=int, required=False,
help="Early stopping patience")
parser.add_argument('--earlystop_min_delta', type=float, required=False, default=1.,
help="Early stopping min percentage delta")
# POCS
parser.add_argument('--pocs_alpha', type=float, required=False, default=0.1,
help='POCS data weighting.')
parser.add_argument('--pocs_thresh', type=float, required=False, default=5.,
help='POCS thresholding percentage')
parser.add_argument('--pocs_weight', type=float, required=False,
help='POCS regularization weight')
args = parser.parse_args()
if args.upsample == "linear":
args.upsample = "trilinear" if args.datadim == "3d" else "bilinear"
if args.patch_shape is None:
if args.datadim == '2d':
args.patch_shape = [-1, -1]
else:
args.patch_shape = [-1, -1, -1]
if args.patch_stride is None:
args.patch_stride = args.patch_shape
if args.earlystop_patience is None:
args.earlystop_patience = args.epochs
# if len(args.netdir) != 0:
# args.net = "load"
return args
def net_args_are_same(args1: Namespace, args2: Namespace) -> bool:
keys_must = ["datadim",
"slice",
"imgchannel",
"patch_shape",
"inputdepth",
"loss",
"lr",
"lr_factor",
"lr_thresh",
"lr_patience",
"reduce_lr",
]
keys_mild = ["net",
"activation",
"last_activation",
"dropout",
"filters",
"skip",
"upsample",
"inittype",
"initgain",
]
errors = []
warnings = []
for k in keys_must:
if args1.__dict__[k] != args2.__dict__[k]:
errors.append(k)
for k in keys_mild:
if args1.__dict__[k] != args2.__dict__[k]:
warnings.append(k)
if len(errors) != 0:
print("The following arguments keys have to be the same:\n\t")
print(", ".join(errors))
return False
if len(warnings) != 0:
print("\nThe following arguments are different, but they are overridden by the network loading:")
print("\t", ", ".join(warnings))
return True