-
Notifications
You must be signed in to change notification settings - Fork 517
/
resnext.py
executable file
·167 lines (125 loc) · 6.64 KB
/
resnext.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
"""ResNeXt in PyTorch.
See the paper "Aggregated Residual Transformations for Deep Neural Networks" for more details.
Code adapted from https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py
"""
from typing import Optional, Callable
import torch.nn as nn
import torch.nn.functional as F
from super_gradients.common.registry.registry import register_model
from super_gradients.common.object_names import Models
from super_gradients.training.models import BaseClassifier
def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
"""3x3 convolution with padding"""
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=dilation, groups=groups, bias=False, dilation=dilation)
def conv1x1(in_planes, out_planes, stride=1):
"""1x1 convolution"""
return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
class GroupedConvBlock(nn.Module):
"""Grouped convolution block."""
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1, base_width=64, dilation=1, norm_layer=None):
super(GroupedConvBlock, self).__init__()
if norm_layer is None:
norm_layer = nn.BatchNorm2d
self.norm_layer = norm_layer
width = int(planes * (base_width / 64.0)) * groups
# Both self.conv2 and self.downsample layers downsample the input when stride != 1
self.conv1 = conv1x1(inplanes, width)
self.bn1 = norm_layer(width)
self.conv2 = conv3x3(width, width, stride, groups, dilation)
self.bn2 = norm_layer(width)
self.conv3 = conv1x1(width, planes * self.expansion)
self.bn3 = norm_layer(planes * self.expansion)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def forward(self, x):
identity = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
if self.downsample is not None:
identity = self.downsample(x)
out += identity
out = self.relu(out)
return out
class ResNeXt(BaseClassifier):
def __init__(self, layers, cardinality, bottleneck_width, num_classes=10, replace_stride_with_dilation=None, in_channels: int = 3):
super(ResNeXt, self).__init__()
if replace_stride_with_dilation is None:
# each element in the tuple indicates if we should replace
# the 2x2 stride with a dilated convolution instead
replace_stride_with_dilation = [False, False, False]
if len(replace_stride_with_dilation) != 3:
raise ValueError("replace_stride_with_dilation should be None " "or a 3-element tuple, got {}".format(replace_stride_with_dilation))
self.cardinality = cardinality
self.dilation = 1
self.inplanes = 64
self.base_width = bottleneck_width
self.conv1 = nn.Conv2d(in_channels, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False)
self.bn1 = nn.BatchNorm2d(self.inplanes)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(GroupedConvBlock, 64, layers[0])
self.layer2 = self._make_layer(GroupedConvBlock, 128, layers[1], stride=2, dilate=replace_stride_with_dilation[0])
self.layer3 = self._make_layer(GroupedConvBlock, 256, layers[2], stride=2, dilate=replace_stride_with_dilation[1])
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
if len(layers) == 4:
self.layer4 = self._make_layer(GroupedConvBlock, 512, layers[3], stride=2, dilate=replace_stride_with_dilation[2])
end_width = 512 if len(layers) == 4 else 256
self.fc = nn.Linear(end_width * GroupedConvBlock.expansion, num_classes)
def _make_layer(self, block, planes, blocks, stride=1, dilate=False):
norm_layer = nn.BatchNorm2d
downsample = None
previous_dilation = self.dilation
if dilate:
self.dilation *= stride
stride = 1
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
conv1x1(self.inplanes, planes * block.expansion, stride),
norm_layer(planes * block.expansion),
)
layers = [block(self.inplanes, planes, stride, downsample, self.cardinality, self.base_width, previous_dilation, norm_layer)]
self.inplanes = planes * block.expansion
for _ in range(1, blocks):
layers.append(block(self.inplanes, planes, groups=self.cardinality, base_width=self.base_width, dilation=self.dilation, norm_layer=norm_layer))
return nn.Sequential(*layers)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = self.maxpool(out)
out = self.layer1(out)
out = self.layer2(out)
out = self.layer3(out)
if self.layer4 is not None:
out = self.layer4(out)
out = self.avgpool(out)
out = out.view(out.size(0), -1)
out = self.fc(out)
return out
def replace_input_channels(self, in_channels: int, compute_new_weights_fn: Optional[Callable[[nn.Module, int], nn.Module]] = None):
from super_gradients.modules.weight_replacement_utils import replace_conv2d_input_channels
self.conv1 = replace_conv2d_input_channels(conv=self.conv1, in_channels=in_channels, fn=compute_new_weights_fn)
def get_input_channels(self) -> int:
return self.conv1.in_channels
class CustomizedResNeXt(ResNeXt):
def __init__(self, arch_params):
super(CustomizedResNeXt, self).__init__(
layers=arch_params.structure if hasattr(arch_params, "structure") else [3, 3, 3],
bottleneck_width=arch_params.num_init_features if hasattr(arch_params, "bottleneck_width") else 64,
cardinality=arch_params.bn_size if hasattr(arch_params, "cardinality") else 32,
num_classes=arch_params.num_classes,
replace_stride_with_dilation=arch_params.replace_stride_with_dilation if hasattr(arch_params, "replace_stride_with_dilation") else None,
)
@register_model(Models.RESNEXT50)
class ResNeXt50(ResNeXt):
def __init__(self, arch_params):
super(ResNeXt50, self).__init__(layers=[3, 4, 6, 3], cardinality=32, bottleneck_width=4, num_classes=arch_params.num_classes)
@register_model(Models.RESNEXT101)
class ResNeXt101(ResNeXt):
def __init__(self, arch_params):
super(ResNeXt101, self).__init__(layers=[3, 4, 23, 3], cardinality=32, bottleneck_width=8, num_classes=arch_params.num_classes)