-
Notifications
You must be signed in to change notification settings - Fork 9
/
cspresnet.py
315 lines (283 loc) · 10 KB
/
cspresnet.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
import paddle
import paddle.nn as nn
import paddle.nn.functional as F
from paddle import ParamAttr
from paddle.regularizer import L2Decay
from paddle.nn.initializer import Constant
from paddleseg.models import layers
from paddleseg.utils import load_entire_model
from ..common_model import swish
from ..builder import BACKBONES
class ConvModule(nn.Layer):
def __init__(self,
ch_in,
ch_out,
filter_size=3,
stride=1,
groups=1,
padding=0,
act=None):
super(ConvModule, self).__init__()
self.conv = nn.Conv2D(
in_channels=ch_in,
out_channels=ch_out,
kernel_size=filter_size,
stride=stride,
padding=padding,
groups=groups,
bias_attr=False)
self.bn = nn.BatchNorm2D(
ch_out,
weight_attr=ParamAttr(regularizer=L2Decay(0.0)),
bias_attr=ParamAttr(regularizer=L2Decay(0.0)))
if act == 'Swish':
self.act = swish()
else:
self.act = layers.Activation(act = act) if act is None or isinstance(act, (
str, dict)) else act
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
x = self.act(x)
return x
class RepVggBlock(nn.Layer):
def __init__(self, ch_in, ch_out, act='relu', alpha=False):
super(RepVggBlock, self).__init__()
self.ch_in = ch_in
self.ch_out = ch_out
self.conv1 = ConvModule(
ch_in, ch_out, 3, stride=1, padding=1, act=None)
self.conv2 = ConvModule(
ch_in, ch_out, 1, stride=1, padding=0, act=None)
if act == 'Swish':
self.act = swish()
else:
self.act = layers.Activation(act = act) if act is None or isinstance(act, (
str, dict)) else act
if alpha:
self.alpha = self.create_parameter(
shape=[1],
attr=ParamAttr(initializer=Constant(value=1.)),
dtype="float32")
else:
self.alpha = None
def forward(self, x):
if hasattr(self, 'conv'):
y = self.conv(x)
else:
if self.alpha:
y = self.conv1(x) + self.alpha * self.conv2(x)
else:
y = self.conv1(x) + self.conv2(x)
y = self.act(y)
return y
def convert_to_deploy(self):
if not hasattr(self, 'conv'):
self.conv = nn.Conv2D(
in_channels=self.ch_in,
out_channels=self.ch_out,
kernel_size=3,
stride=1,
padding=1,
groups=1)
kernel, bias = self.get_equivalent_kernel_bias()
self.conv.weight.set_value(kernel)
self.conv.bias.set_value(bias)
self.__delattr__('conv1')
self.__delattr__('conv2')
def get_equivalent_kernel_bias(self):
kernel3x3, bias3x3 = self._fuse_bn_tensor(self.conv1)
kernel1x1, bias1x1 = self._fuse_bn_tensor(self.conv2)
if self.alpha:
return kernel3x3 + self.alpha * self._pad_1x1_to_3x3_tensor(
kernel1x1), bias3x3 + self.alpha * bias1x1
else:
return kernel3x3 + self._pad_1x1_to_3x3_tensor(
kernel1x1), bias3x3 + bias1x1
def _pad_1x1_to_3x3_tensor(self, kernel1x1):
if kernel1x1 is None:
return 0
else:
return nn.functional.pad(kernel1x1, [1, 1, 1, 1])
def _fuse_bn_tensor(self, branch):
if branch is None:
return 0, 0
kernel = branch.conv.weight
running_mean = branch.bn._mean
running_var = branch.bn._variance
gamma = branch.bn.weight
beta = branch.bn.bias
eps = branch.bn._epsilon
std = (running_var + eps).sqrt()
t = (gamma / std).reshape((-1, 1, 1, 1))
return kernel * t, beta - running_mean * gamma / std
class BasicBlock(nn.Layer):
def __init__(self,
ch_in,
ch_out,
act='relu',
shortcut=True,
use_alpha=False):
super(BasicBlock, self).__init__()
assert ch_in == ch_out
self.conv1 = ConvModule(ch_in, ch_out, 3, stride=1, padding=1, act=act)
self.conv2 = RepVggBlock(ch_out, ch_out, act=act, alpha=use_alpha)
self.shortcut = shortcut
def forward(self, x):
y = self.conv1(x)
y = self.conv2(y)
if self.shortcut:
return paddle.add(x, y)
else:
return y
class EffectiveSELayer(nn.Layer):
""" Effective Squeeze-Excitation
From `CenterMask : Real-Time Anchor-Free Instance Segmentation` - https://arxiv.org/abs/1911.06667
"""
def __init__(self, channels, act='hardsigmoid'):
super(EffectiveSELayer, self).__init__()
self.fc = nn.Conv2D(channels, channels, kernel_size=1, padding=0)
if act == 'Swish':
self.act = swish()
else:
self.act = layers.Activation(act = act) if act is None or isinstance(act, (
str, dict)) else act
def forward(self, x):
x_se = x.mean((2, 3), keepdim=True)
x_se = self.fc(x_se)
return x * self.act(x_se)
class CSPResStage(nn.Layer):
def __init__(self,
block_fn,
ch_in,
ch_out,
n,
stride,
act='relu',
attn='eca',
use_alpha=False):
super(CSPResStage, self).__init__()
ch_mid = (ch_in + ch_out) // 2
if stride == 2:
self.conv_down = ConvModule(
ch_in, ch_mid, 3, stride=2, padding=1, act=act)
else:
self.conv_down = None
self.conv1 = ConvModule(ch_mid, ch_mid // 2, 1, act=act)
self.conv2 = ConvModule(ch_mid, ch_mid // 2, 1, act=act)
self.blocks = nn.Sequential(*[
block_fn(
ch_mid // 2,
ch_mid // 2,
act=act,
shortcut=True,
use_alpha=use_alpha) for i in range(n)
])
if attn:
self.attn = EffectiveSELayer(ch_mid, act='hardsigmoid')
else:
self.attn = None
self.conv3 = ConvModule(ch_mid, ch_out, 1, act=act)
def forward(self, x):
if self.conv_down is not None:
x = self.conv_down(x)
y1 = self.conv1(x)
y2 = self.blocks(self.conv2(x))
y = paddle.concat([y1, y2], axis=1)
if self.attn is not None:
y = self.attn(y)
y = self.conv3(y)
return y
@BACKBONES.register()
class CSPResNet(nn.Layer):
__shared__ = ['width_mult', 'depth_mult', 'trt']
def __init__(self,
layers_=[3, 6, 6, 3],
channels=[64, 128, 256, 512, 1024],
act='Swish',
return_idx=[1, 2, 3],
depth_wise=False,
use_large_stem=False,
width_mult=1.0,
depth_mult=1.0,
trt=False,
use_checkpoint=False,
use_alpha=False,
out_conv = False,
pretrain = None,
cfg = None,
**args):
super(CSPResNet, self).__init__()
self.use_checkpoint = use_checkpoint
channels = [max(round(c * width_mult), 1) for c in channels]
layers_ = [max(round(l * depth_mult), 1) for l in layers_]
if act == 'Swish':
self.act = swish()
else:
self.act = layers.Activation(act = act) if act is None or isinstance(act, (
str, dict)) else act
if use_large_stem:
self.stem = nn.Sequential(
('conv1', ConvModule(
3, channels[0] // 2, 3, stride=2, padding=1, act=act)),
('conv2', ConvModule(
channels[0] // 2,
channels[0] // 2,
3,
stride=1,
padding=1,
act=act)), ('conv3', ConvModule(
channels[0] // 2,
channels[0],
3,
stride=1,
padding=1,
act=act)))
else:
self.stem = nn.Sequential(
('conv1', ConvModule(
3, channels[0] // 2, 3, stride=2, padding=1, act=act)),
('conv2', ConvModule(
channels[0] // 2,
channels[0],
3,
stride=1,
padding=1,
act=act)))
n = len(channels) - 1
self.stages = nn.Sequential(*[(str(i), CSPResStage(
BasicBlock,
channels[i],
channels[i + 1],
layers_[i],
2,
act=act,
use_alpha=use_alpha)) for i in range(n)])
self._out_channels = channels[1:]
self._out_strides = [4 * 2**i for i in range(n)]
self.return_idx = return_idx
if use_checkpoint:
paddle.seed(0)
if out_conv:
self.out_conv = nn.Conv2D(channels[-1],cfg.featuremap_out_channel,1)
if pretrain is not None:
self.load_pretrain(pretrain)
def load_pretrain(self,pretrain):
# for name,param in self.named_parameters():
# print(name)
load_entire_model(self,pretrain)
def forward(self, inputs):
x = inputs
x = self.stem(x)
outs = []
for idx, stage in enumerate(self.stages):
if self.use_checkpoint and self.training:
x = paddle.distributed.fleet.utils.recompute(
stage, x, **{"preserve_rng_state": True})
else:
x = stage(x)
if idx in self.return_idx:
outs.append(x)
if hasattr(self,'out_conv'):
outs[-1] = self.out_conv(outs[-1])
return outs