-
Notifications
You must be signed in to change notification settings - Fork 0
/
weight_batch_norm.py
298 lines (236 loc) · 15.4 KB
/
weight_batch_norm.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
from torch.nn.modules.batchnorm import _BatchNorm
from torch.nn.parameter import Parameter
from torch.nn import functional as F
import torch
from torch.nn import init
from torch.nn.modules import Module
from torch import nn
class _NormBase2(Module):
"""Common base of _InstanceNorm and _BatchNorm"""
_version = 2
__constants__ = ['track_running_stats', 'momentum', 'eps', 'weight', 'bias',
'running_mean', 'running_var', 'num_batches_tracked',
'num_features', 'affine']
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True,
track_running_stats=True):
super(_NormBase2, self).__init__()
self.num_features = num_features
self.eps = eps
self.momentum = momentum
self.affine = affine
self.track_running_stats = track_running_stats
if self.affine:
self.weight = Parameter(torch.Tensor(num_features))
self.bias = Parameter(torch.Tensor(num_features))
else:
self.register_parameter('weight', None)
self.register_parameter('bias', None)
if self.track_running_stats:
self.register_buffer('running_mean', torch.zeros(num_features))
self.register_buffer('running_var', torch.ones(num_features))
self.register_buffer('num_batches_tracked', torch.tensor(0, dtype=torch.long))
else:
self.register_parameter('running_mean', None)
self.register_parameter('running_var', None)
self.register_parameter('num_batches_tracked', None)
self.reset_parameters()
def reset_running_stats(self):
if self.track_running_stats:
self.running_mean.zero_()
self.running_var.fill_(1)
self.num_batches_tracked.zero_()
def reset_parameters(self):
self.reset_running_stats()
if self.affine:
init.ones_(self.weight)
init.zeros_(self.bias)
def _check_input_dim(self, input):
raise NotImplementedError
def extra_repr(self):
return '{num_features}, eps={eps}, momentum={momentum}, affine={affine}, ' \
'track_running_stats={track_running_stats}'.format(**self.__dict__)
def _load_from_state_dict(self, state_dict, prefix, local_metadata, strict,
missing_keys, unexpected_keys, error_msgs):
version = local_metadata.get('version', None)
if (version is None or version < 2) and self.track_running_stats:
# at version 2: added num_batches_tracked buffer
# this should have a default value of 0
num_batches_tracked_key = prefix + 'num_batches_tracked'
if num_batches_tracked_key not in state_dict:
state_dict[num_batches_tracked_key] = torch.tensor(0, dtype=torch.long)
super(_NormBase2, self)._load_from_state_dict(
state_dict, prefix, local_metadata, strict,
missing_keys, unexpected_keys, error_msgs)
class Batch_test(_BatchNorm):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(Batch_test, self).__init__(num_features, eps, momentum, affine)
def forward(self, input):
if self.training:
# self.running_mean.mul_(1 - self.momentum).add_(input.mean([0, 2, 3]).detach() * self.momentum)
# self.running_var.mul_(1 - self.momentum).add_(input.var([0, 2, 3]).detach() * self.momentum)
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - input.mean([0, 2, 3]).view([1, self.num_features, 1, 1]).expand_as(input)) / (
input.var([0, 2, 3], unbiased=False).view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view(
[1, self.num_features, 1, 1]).expand_as(input)
else:
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - self.running_mean.view([1, self.num_features, 1, 1]).expand_as(input)) / (
self.running_var.view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features, 1, 1]).expand_as(input)
class Batch_test2(_BatchNorm):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(Batch_test2, self).__init__(num_features, eps, momentum, affine)
def forward(self, input):
if self.training:
# self.running_mean.mul_(1 - self.momentum).add_(input.mean([0, 2, 3]).detach() * self.momentum)
# self.running_var.mul_(1 - self.momentum).add_(input.var([0, 2, 3]).detach() * self.momentum)
return F.batch_norm(input, None, None, self.weight, self.bias, True, self.momentum, self.eps)
else:
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - self.running_mean.view([1, self.num_features, 1, 1]).expand_as(input)) / (
self.running_var.view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features, 1, 1]).expand_as(input)
class _BatchNorm2d(_BatchNorm):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(_BatchNorm2d, self).__init__(num_features, eps, momentum, affine)
def forward(self, input):
self._check_input_dim(input)
if self.training:
self.running_mean.mul_(1 - self.momentum).add_(input.mean([0, 2, 3]).detach() * self.momentum)
self.running_var.mul_(1 - self.momentum).add_(input.var([0, 2, 3]).detach() * self.momentum)
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - input.mean([0, 2, 3]).view([1, self.num_features, 1, 1]).expand_as(input)) / (
input.var([0, 2, 3], unbiased=False).view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view(
[1, self.num_features, 1, 1]).expand_as(input)
else:
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - self.running_mean.view([1, self.num_features, 1, 1]).expand_as(input)) / (
self.running_var.view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features, 1, 1]).expand_as(input)
class _BatchNorm2d_ntest(_BatchNorm):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(_BatchNorm2d_ntest, self).__init__(num_features, eps, momentum, affine)
def forward(self, input):
# self._check_input_dim(input)
if self.training:
self.running_mean.mul_(1 - self.momentum).add_(input.mean([0, 2, 3]).detach() * self.momentum)
self.running_var.mul_(1 - self.momentum).add_(input.var([0, 2, 3]).detach() * self.momentum)
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - input.mean([0, 2, 3]).view([1, self.num_features, 1, 1]).expand_as(input)) / (
input.var([0, 2, 3], unbiased=False).view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view(
[1, self.num_features, 1, 1]).expand_as(input)
else:
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - input.mean([0, 2, 3]).view([1, self.num_features, 1, 1]).expand_as(input)) / (
input.var([0, 2, 3], unbiased=False).view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view(
[1, self.num_features, 1, 1]).expand_as(input)
class BatchNorm2d(_BatchNorm2d):
def _check_input_dim(self, input):
if input.dim() != 4:
raise ValueError('expected 4D input (got {}D input)'.format(input.dim()))
class _BatchNorm1d(_BatchNorm):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(_BatchNorm1d, self).__init__(num_features, eps, momentum, affine)
def forward(self, input):
self._check_input_dim(input)
if self.training:
self.running_mean.mul_(1 - self.momentum).add_(input.mean([0]).detach() * self.momentum)
self.running_var.mul_(1 - self.momentum).add_(input.var([0]).detach() * self.momentum)
return self.weight.view([1, self.num_features]).expand_as(input) * (input - input.mean([0]).view([1, self.num_features]).expand_as(input)) / (
input.var([0], unbiased=False).view([1, self.num_features]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features]).expand_as(input)
else:
return self.weight.view([1, self.num_features]).expand_as(input) * (input - self.running_mean.view([1, self.num_features]).expand_as(input)) / (
self.running_var.view([1, self.num_features]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features]).expand_as(input)
class BatchNorm1d(_BatchNorm1d):
def _check_input_dim(self, input):
if input.dim() != 2:
raise ValueError('expected 2D (got {}D input)'.format(input.dim()))
class _BatchWeightNorm1d(_NormBase2):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(_BatchWeightNorm1d, self).__init__(num_features, eps, momentum, affine)
self.update_batch_stats = True
def forward(self, input, w):
self._check_input_dim(input)
if self.training:
w_exd = w.view([w.shape[0], 1]).expand_as(input)
mean = (w_exd * input).mean([0])
mean = mean * input.shape[0] / w.sum()
var_, var = get_var1d(input, mean, w)
if self.update_batch_stats:
self.running_mean.mul_(1 - self.momentum).add_(mean.detach() * self.momentum)
self.running_var.mul_(1 - self.momentum).add_(var.detach() * self.momentum)
return self.weight.view([1, self.num_features]).expand_as(input) * (input - mean.view([1, self.num_features]).expand_as(input)) / (
var_.view([1, self.num_features]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features]).expand_as(input)
else:
return self.weight.view([1, self.num_features]).expand_as(input) * (input - self.running_mean.view([1, self.num_features]).expand_as(input)) / (
self.running_var.view([1, self.num_features]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features]).expand_as(input)
def reset_parameters(self):
nn.init.constant_(self.weight, 1)
nn.init.constant_(self.bias, 0)
self.running_mean.zero_()
self.running_var.fill_(1)
class BatchWeightNorm1d(_BatchWeightNorm1d):
def _check_input_dim(self, input):
if input.dim() != 2:
raise ValueError('expected 2D (got {}D input)'.format(input.dim()))
class _BatchWeightNorm2d(_NormBase2):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(_BatchWeightNorm2d, self).__init__(num_features, eps, momentum, affine)
self.update_batch_stats = True
def forward(self, input, w):
self._check_input_dim(input)
if self.training:
w_exd = w.view([w.shape[0], 1, 1, 1]).expand_as(input)
mean = (w_exd * input).mean([0, 2, 3])
mean = mean * input.shape[0] / w.sum()
var_, var = get_var2d(input, mean, w)
if self.update_batch_stats:
self.running_mean.mul_(1 - self.momentum).add_(mean.detach() * self.momentum)
self.running_var.mul_(1 - self.momentum).add_(var.detach() * self.momentum)
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - mean.view([1, self.num_features, 1, 1]).expand_as(input)) / (
var_.view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features, 1, 1]).expand_as(input)
else:
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - self.running_mean.view([1, self.num_features, 1, 1]).expand_as(input)) / (
self.running_var.view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features, 1, 1]).expand_as(input)
def reset_parameters(self):
nn.init.constant_(self.weight, 1)
nn.init.constant_(self.bias, 0)
self.running_mean.zero_()
self.running_var.fill_(1)
class _BatchWeightNorm2d_ntest(_NormBase2):
def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
super(_BatchWeightNorm2d_ntest, self).__init__(num_features, eps, momentum, affine)
self.update_batch_stats = True
def forward(self, input, w):
if self.training:
w_exd = w.view([w.shape[0], 1, 1, 1]).expand_as(input)
mean = (w_exd * input).mean([0, 2, 3])
mean = mean * input.shape[0] / w.sum()
var_, var = get_var2d(input, mean, w)
if self.update_batch_stats:
self.running_mean.mul_(1 - self.momentum).add_(mean.detach() * self.momentum)
self.running_var.mul_(1 - self.momentum).add_(var.detach() * self.momentum)
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - mean.view([1, self.num_features, 1, 1]).expand_as(input)) / (
var_.view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features, 1, 1]).expand_as(input)
else:
w_exd = w.view([w.shape[0], 1, 1, 1]).expand_as(input)
mean = (w_exd * input).mean([0, 2, 3])
mean = mean * input.shape[0] / w.sum()
var_, var = get_var2d(input, mean, w)
return self.weight.view([1, self.num_features, 1, 1]).expand_as(input) * (input - mean.view([1, self.num_features, 1, 1]).expand_as(input)) / (
var_.view([1, self.num_features, 1, 1]).expand_as(input) + self.eps).sqrt() + self.bias.view([1, self.num_features, 1, 1]).expand_as(input)
def reset_parameters(self):
nn.init.constant_(self.weight, 1)
nn.init.constant_(self.bias, 0)
self.running_mean.zero_()
self.running_var.fill_(1)
class BatchWeightNorm2d(_BatchWeightNorm2d):
def _check_input_dim(self, input):
if input.dim() != 4:
raise ValueError('expected 4D input (got {}D input)'.format(input.dim()))
def get_var2d(x, new_mean, w):
scale = x.size(0) * x.size(2) * x.size(3) * w.sum() / x.size(0)
new_mean = new_mean.view([1, new_mean.shape[0], 1, 1]).expand_as(x)
tmp = torch.pow((x - new_mean), 2)
tem_v = tmp * w.view([w.shape[0], 1, 1, 1]).expand_as(tmp)
new_var = tem_v.mean([0, 2, 3]) * x.size(0) / w.sum()
runing_temp_var = new_var * scale / (scale - 1)
return new_var, runing_temp_var
def get_var1d(x, new_mean, w):
scale = x.size(0) * w.sum() / x.size(0)
new_mean = new_mean.view([1, new_mean.shape[0]]).expand_as(x)
tmp = torch.pow((x - new_mean), 2)
tem_v = tmp * w.view([w.shape[0], 1]).expand_as(tmp)
new_var = tem_v.mean([0]) * x.size(0) / w.sum()
runing_temp_var = new_var * scale / (scale - 1)
return new_var, runing_temp_var