ResNet

字数统计: 825字   |   阅读时长: 5分

将传统的ResNet算法改成基于inplace-abn的算法，可以明显减少显存占用和训练时间，将代码记录下来以便以后使用。

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
import torch
from torch import nn
from inplace_abn import InPlaceABN


def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
    return nn.Conv2d(in_planes, out_planes, (3, 3), (stride, stride), (dilation, dilation), groups=groups, bias=False, dilation=(dilation, dilation))


def conv1x1(in_planes, out_planes, stride=1):
    return nn.Conv2d(in_planes, out_planes, (1, 1), (stride, stride), bias=False)


class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, in_planes, planes, stride=1, downsample=None, groups=1, base_width=64, dilation=1, abn=InPlaceABN):
        super(BasicBlock, self).__init__()
        if groups != 1 or base_width != 64:
            raise ValueError('BasicBlock only supports groups=1 and base_width=64')
        if dilation > 1:
            raise NotImplementedError("Dilation > 1 not supported in BasicBlock")

        # Both self.conv1 and self.downsample layers downsample the input when stride != 1
        self.conv1 = conv3x3(in_planes, planes, stride)
        self.bn_relu = abn(planes)

        self.conv2 = conv3x3(planes, planes)
        self.bn = abn(planes, activation="identity")
        self.relu = nn.LeakyReLU(inplace=True)

        self.downsample = downsample
        self.stride = stride

    def forward(self, x: torch.Tensor):
        identity = x

        out = self.conv1(x)
        out = self.bn_relu(out)

        out = self.conv2(out)
        out = self.bn(out)

        if self.downsample is not None:
            identity = self.downsample(x)

        return self.relu(out + identity)


class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, in_planes, planes, stride=1, downsample=None, groups=1, base_width=64, dilation=1, abn=InPlaceABN):
        super(Bottleneck, self).__init__()
        width = int(planes * (base_width / 64.)) * groups
        # Both self.conv2 and self.downsample layers downsample the input when stride != 1
        self.conv1 = conv1x1(in_planes, width)
        self.bn_relu1 = abn(width)

        self.conv2 = conv3x3(width, width, stride, groups, dilation)
        self.bn_relu2 = abn(width)

        self.conv3 = conv1x1(width, planes * self.expansion)
        self.bn3 = abn(planes * self.expansion, activation="identity")
        self.relu = nn.LeakyReLU(inplace=True)

        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn_relu1(out)

        out = self.conv2(out)
        out = self.bn_relu2(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            identity = self.downsample(x)

        return self.relu(out + identity)


class ResNet(nn.Module):

    def __init__(self, block, layers, num_classes=1000, groups=1, width_per_group=64, strides=(1, 2, 2, 2), dilation=(False, False, False, False), abn=InPlaceABN, return_features=False):
        super(ResNet, self).__init__()
        self._abn = abn
        self.return_features = return_features

        self.in_planes = 64
        self.dilation = 1
        self.groups = groups
        self.base_width = width_per_group

        self.conv1 = nn.Sequential(nn.Conv2d(3, self.in_planes, (7, 7), (2, 2), (3, 3), bias=False), abn(self.in_planes))
        self.max_pool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)

        self.layer1 = self._make_layer(block, 64, layers[0], stride=strides[0], dilate=dilation[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=strides[1], dilate=dilation[1])
        self.layer3 = self._make_layer(block, 256, layers[2], stride=strides[2], dilate=dilation[2])
        self.layer4 = self._make_layer(block, 512, layers[3], stride=strides[3], dilate=dilation[3])
        self.avg_pool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(512 * block.expansion, num_classes)

    def _make_layer(self, block, planes, blocks, stride=1, dilate=False):
        abn = self._abn
        downsample = None
        previous_dilation = self.dilation
        if dilate:
            self.dilation *= stride
            stride = 1

        if stride != 1 or self.in_planes != planes * block.expansion:
            downsample = nn.Sequential(conv1x1(self.in_planes, planes * block.expansion, stride), abn(planes * block.expansion, activation="identity"), )

        layers = []
        layers.append(block(self.in_planes, planes, stride, downsample, self.groups, self.base_width, previous_dilation, abn))
        self.in_planes = planes * block.expansion
        for _ in range(1, blocks):
            layers.append(block(self.in_planes, planes, groups=self.groups, base_width=self.base_width, dilation=self.dilation, abn=abn))

        return nn.Sequential(*layers)

    def _forward_impl(self, x):
        # See note [TorchScript super()]
        x1 = x = self.conv1(x)
        x = self.max_pool(x)

        x2 = x = self.layer1(x)
        x3 = x = self.layer2(x)
        x4 = x = self.layer3(x)
        x5 = x = self.layer4(x)

        if self.return_features:
            return [x1, x2, x3, x4, x5]

        x = self.avg_pool(x)
        x = torch.flatten(x, 1)
        return self.fc(x)

    def forward(self, x):
        return self._forward_impl(x)


def _resnet(arch, block, layers, pretrained, progress, **kwargs) -> ResNet:
    return ResNet(block, layers, **kwargs)


def resnet18(pretrained=False, progress=True, **kwargs) -> ResNet:
    return _resnet('resnet18', BasicBlock, [2, 2, 2, 2], pretrained, progress, **kwargs)


def resnet34(pretrained=False, progress=True, **kwargs) -> ResNet:
    return _resnet('resnet34', BasicBlock, [3, 4, 6, 3], pretrained, progress, **kwargs)


def resnet50(pretrained=False, progress=True, **kwargs) -> ResNet:
    return _resnet('resnet50', Bottleneck, [3, 4, 6, 3], pretrained, progress, **kwargs)


def resnet101(pretrained=False, progress=True, **kwargs) -> ResNet:
    return _resnet('resnet101', Bottleneck, [3, 4, 23, 3], pretrained, progress, **kwargs)


def resnet152(pretrained=False, progress=True, **kwargs) -> ResNet:
    return _resnet('resnet152', Bottleneck, [3, 8, 36, 3], pretrained, progress, **kwargs)
赏

非常感谢,我会继续努力的！

支付宝
微信

扫一扫，分享到微信

tag:

    缺失模块。
    1、请确保node版本大于6.2
    2、在博客根目录（注意不是yilia-plus根目录）执行以下命令：
    npm i hexo-generator-json-content --save

    3、在根目录_config.yml里添加配置： 

      jsonContent:
    meta: false
    pages: false
    posts:
      title: true
      date: true
      path: true
      text: false
      raw: false
      content: false
      slug: false
      updated: false
      comments: false
      link: false
      permalink: false
      excerpt: false
      categories: false
      tags: true

真正使人进步的,是不断总结。