-
Notifications
You must be signed in to change notification settings - Fork 2
/
blocks.py
303 lines (250 loc) · 11.3 KB
/
blocks.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.init as init
from plif_node import NEURON
from torch.autograd import Variable
PRIMITIVES = {
'skip_connect': lambda C_in, C_out, stride, snn_params: Identity(C_in, C_out, snn_params) if stride == 1 else FactorizedReduce(C_in, C_out, snn_params),
'max_pool_k2' : lambda C_in, C_out, stride, snn_params: SpikingMaxPool2d(2, stride, snn_params),
'avg_pool_k2' : lambda C_in, C_out, stride, snn_params: SpikingAvgPool2d(2, stride, snn_params),
'SCB_k3': lambda C_in, C_out, stride, snn_params: SpikingConvBlock(C_in, C_out, 3, stride, snn_params),
'SCB_k5': lambda C_in, C_out, stride, snn_params: SpikingConvBlock(C_in, C_out, 5, stride, snn_params),
'SCB_k7': lambda C_in, C_out, stride, snn_params: SpikingConvBlock(C_in, C_out, 7, stride, snn_params),
'SRB_k3': lambda C_in, C_out, stride, snn_params: SpikingResidualBlock(C_in, C_out, 3, stride, snn_params),
'SRB_k5': lambda C_in, C_out, stride, snn_params: SpikingResidualBlock(C_in, C_out, 5, stride, snn_params),
'SRB_k7': lambda C_in, C_out, stride, snn_params: SpikingResidualBlock(C_in, C_out, 7, stride, snn_params),
'SIB_k3_e1': lambda C_in, C_out, stride, snn_params: SpikingInvertedBottleneck(C_in, C_out, 3, 1, stride, snn_params),
'SIB_k3_e3': lambda C_in, C_out, stride, snn_params: SpikingInvertedBottleneck(C_in, C_out, 3, 3, stride, snn_params),
'SIB_k3_e6': lambda C_in, C_out, stride, snn_params: SpikingInvertedBottleneck(C_in, C_out, 3, 6, stride, snn_params),
'SIB_k5_e1': lambda C_in, C_out, stride, snn_params: SpikingInvertedBottleneck(C_in, C_out, 5, 1, stride, snn_params),
'SIB_k5_e3': lambda C_in, C_out, stride, snn_params: SpikingInvertedBottleneck(C_in, C_out, 5, 3, stride, snn_params),
'SIB_k5_e6': lambda C_in, C_out, stride, snn_params: SpikingInvertedBottleneck(C_in, C_out, 5, 6, stride, snn_params),
}
class Identity(nn.Module):
def __init__(self, C_in, C_out, snn_params=None):
super(Identity, self).__init__()
self.identity = C_in == C_out
if not self.identity:
# SNN parameters
if snn_params is not None:
init_tau = snn_params['init_tau']
v_threshold = snn_params['v_threshold']
neuron = snn_params['neuron']
self.conv = nn.Conv2d(C_in, C_out, 1, stride=1, padding=0, bias=False)
self.bn = nn.BatchNorm2d(C_out)
self.spike_neuron = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
def forward(self, x):
if self.identity:
return x, 0
else:
spikes = 0
out = self.bn(self.conv(x))
if self.spike_neuron:
out = self.spike_neuron(out)
spikes += out.sum().item()
else: # for ANN
out = F.relu(out)
return out, spikes
class FactorizedReduce(nn.Module):
def __init__(self, C_in, C_out, snn_params=None):
super(FactorizedReduce, self).__init__()
assert C_out % 2 == 0
# SNN parameters
if snn_params is not None:
init_tau = snn_params['init_tau']
v_threshold = snn_params['v_threshold']
neuron = snn_params['neuron']
self.conv_1 = nn.Conv2d(C_in, C_out // 2, 1, stride=2, padding=0, bias=False)
self.conv_2 = nn.Conv2d(C_in, C_out // 2, 1, stride=2, padding=0, bias=False)
self.bn = nn.BatchNorm2d(C_out)
self.spike_neuron = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
def forward(self, x):
spikes = 0
out = self.bn(torch.cat([self.conv_1(x), self.conv_2(x[:,:,1:,1:])], dim=1))
if self.spike_neuron:
out = self.spike_neuron(out)
spikes += out.sum().item()
else:
out = F.relu(out)
return out, spikes
""" 2D max pooling
"""
class SpikingMaxPool2d(nn.Module):
def __init__(self, kernel_size, stride, snn_params=None):
super(SpikingMaxPool2d, self).__init__()
# SNN parameters
if snn_params is not None:
self.neuron = snn_params['neuron']
self.maxpool = nn.MaxPool2d(kernel_size, stride)
def forward(self, x):
spikes = 0
out = self.maxpool(x)
if not(self.neuron == 'ANN'):
spikes += out.sum().item()
return out, spikes
""" 2D avg pooling
"""
class SpikingAvgPool2d(nn.Module):
def __init__(self, kernel_size, stride, snn_params=None):
super(SpikingAvgPool2d, self).__init__()
# SNN parameters
if snn_params is not None:
init_tau = snn_params['init_tau']
v_threshold = snn_params['v_threshold']
self.neuron = snn_params['neuron']
self.avgpool = nn.AvgPool2d(kernel_size, stride)
self.spike_neuron = None if self.neuron == 'ANN' else NEURON[self.neuron](init_tau, v_threshold)
def forward(self, x):
spikes = 0
out = self.avgpool(x)
if not(self.neuron == 'ANN'):
out = self.spike_neuron(out)
spikes += out.sum().item()
return out, spikes
""" Spiking ConvBlock
- two conv
"""
class SpikingConvBlock(nn.Module):
def __init__(self, inplanes, planes, kernel_size=3, stride = 1, snn_params = None):
super(SpikingConvBlock, self).__init__()
self.stride = stride
# SNN parameters
if snn_params is not None:
init_tau = snn_params['init_tau']
v_threshold = snn_params['v_threshold']
neuron = snn_params['neuron']
# first conv
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=kernel_size, stride=stride, padding=(kernel_size//2), groups=1, bias=False, dilation=1)
self.bn1 = nn.BatchNorm2d(planes)
self.spike_neuron_1 = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
# second conv
self.conv2 = nn.Conv2d(planes, planes, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=1, bias=False, dilation=1)
self.bn2 = nn.BatchNorm2d(planes)
self.spike_neuron_2 = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
def forward(self, x):
spikes = 0
out = self.bn1(self.conv1(x))
if self.spike_neuron_1:
out = self.spike_neuron_1(out)
tmp_spikes = out.sum().item()
spikes += tmp_spikes
else:
out = F.relu(out)
out = self.bn2(self.conv2(out))
if self.spike_neuron_2:
out = self.spike_neuron_2(out)
tmp_spikes = out.sum().item()
spikes += tmp_spikes
else:
out = F.relu(out)
return out, spikes
""" Spiking ResidualBlock
- two conv and skip connection
"""
class SpikingResidualBlock(nn.Module):
def __init__(self, inplanes, planes, kernel_size=3, stride = 1, snn_params = None):
super(SpikingResidualBlock, self).__init__()
self.stride = stride
# SNN parameters
if snn_params is not None:
init_tau = snn_params['init_tau']
v_threshold = snn_params['v_threshold']
neuron = snn_params['neuron']
# first conv
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=kernel_size, stride=stride, padding=(kernel_size//2), groups=1, bias=False, dilation=1)
self.bn1 = nn.BatchNorm2d(planes)
self.spike_neuron_1 = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
# second conv
self.conv2 = nn.Conv2d(planes, planes, kernel_size=kernel_size, stride=1, padding=(kernel_size//2), groups=1, bias=False, dilation=1)
self.bn2 = nn.BatchNorm2d(planes)
self.spike_neuron_2 = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
# skip
self.use_res_connect = stride == 1 and inplanes == planes
if not self.use_res_connect:
self.downsample = nn.Sequential(
nn.Conv2d(inplanes, planes, kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(planes),
)
def forward(self, x):
spikes = 0
out = self.bn1(self.conv1(x))
if self.spike_neuron_1:
out = self.spike_neuron_1(out)
tmp_spikes = out.sum().item()
spikes += tmp_spikes
else:
out = F.relu(out)
out = self.bn2(self.conv2(out))
if self.use_res_connect:
out = out + x
else:
out = out + self.downsample(x)
if self.spike_neuron_2:
out = self.spike_neuron_2(out)
tmp_spikes = out.sum().item()
spikes += tmp_spikes
else:
out = F.relu(out)
return out, spikes
""" Spiking InvertedBottleneck
"""
class SpikingInvertedBottleneck(nn.Module):
def __init__(self, in_planes, out_planes, kernel_size, expansion, stride, snn_params=None):
super(SpikingInvertedBottleneck, self).__init__()
self.stride = stride
# SNN parameters
if snn_params is not None:
init_tau = snn_params['init_tau']
v_threshold = snn_params['v_threshold']
neuron = snn_params['neuron']
# first conv (point-wise)
planes = expansion * in_planes
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=1, stride=1, padding=0, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
self.spike_neuron_1 = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
# second conv (depth-wise)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=kernel_size, stride=stride, padding=(kernel_size//2), bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.spike_neuron_2 = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
# third conv (point-wise)
self.conv3 = nn.Conv2d(planes, out_planes, kernel_size=1, stride=1, padding=0, bias=False)
self.bn3 = nn.BatchNorm2d(out_planes)
self.spike_neuron_3 = None if neuron == 'ANN' else NEURON[neuron](init_tau, v_threshold)
self.use_res_connect = self.stride == 1 and in_planes == out_planes
self.downsample = None
if stride == 1 and in_planes != out_planes:
self.downsample = nn.Sequential(
nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=1, padding=0, bias=False),
nn.BatchNorm2d(out_planes),
)
# when stride == 2, there is no downsample connection (MobileNetV2 paper, Figure 4-(d))
def forward(self, x):
spikes = 0
out = self.bn1(self.conv1(x))
if self.spike_neuron_1:
out = self.spike_neuron_1(out)
tmp_spikes = out.sum().item()
spikes += tmp_spikes
else:
out = F.relu(out)
out = self.bn2(self.conv2(out))
if self.spike_neuron_2:
out = self.spike_neuron_2(out)
tmp_spikes = out.sum().item()
spikes += tmp_spikes
else:
out = F.relu(out)
out = self.bn3(self.conv3(out))
if self.use_res_connect:
out = out + x
else:
if self.downsample is not None:
out = out + self.downsample(x)
if self.spike_neuron_3:
out = self.spike_neuron_3(out)
tmp_spikes = out.sum().item()
spikes += tmp_spikes
else:
out = F.relu(out)
return out, spikes