pytorch_预训练resnext自定义类别数量_源码分享

系列文章目录

9种经典图片分类卷积模型系列合集（推荐程度依次递减）：

1. Se_resnet50
2. Resnet50
3. Xception
4. inceptionresnetv2
5. resnext
6. bninception
7. shufflenetv2
8. polynet
9. vggm

Imagenet的预训练resnext是1000个类别，根据笔者的经验，预训练的网络结构是不能更改的，改了效果就不行，只能在末尾增加一个（2048，classes）的全连接层，中间还有一层dropout，这样效果是最好的。

源码

from __future__ import print_function, division, absolute_import
import os
import torch
import torch.nn as nn
import torch.utils.model_zoo as model_zoo
from .resnext_features import resnext101_32x4d_features
from .resnext_features import resnext101_64x4d_features

__all__ = ['ResNeXt101_32x4d', 'resnext101_32x4d',
           'ResNeXt101_64x4d', 'resnext101_64x4d']

pretrained_settings = {
    'resnext101_32x4d': {
        True: {
            'url': 'http://data.lip6.fr/cadene/pretrainedmodels/resnext101_32x4d-29e315fa.pth',
            'input_space': 'RGB',
            'input_size': [3, 224, 224],
            'input_range': [0, 1],
            'mean': [0.485, 0.456, 0.406],
            'std': [0.229, 0.224, 0.225],
            'num_classes': 1000
        }
    },
    'resnext101_64x4d': {
        True: {
            'url': 'http://data.lip6.fr/cadene/pretrainedmodels/resnext101_64x4d-e77a0586.pth',
            'input_space': 'RGB',
            'input_size': [3, 224, 224],
            'input_range': [0, 1],
            'mean': [0.485, 0.456, 0.406],
            'std': [0.229, 0.224, 0.225],
            'num_classes': 1000
        }
    }
}


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 BasicBlock(nn.Module):
    expansion = 1
    __constants__ = ['downsample']

    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
                 base_width=64, dilation=1, norm_layer=None):
        super(BasicBlock, self).__init__()
        if norm_layer is None:
            norm_layer = nn.BatchNorm2d
        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(inplanes, planes, stride)
        self.bn1 = norm_layer(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = norm_layer(planes)
        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)

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

        out += identity
        out = self.relu(out)

        return out


class Bottleneck(nn.Module):
    expansion = 4
    __constants__ = ['downsample']

    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
                 base_width=64, dilation=1, norm_layer=None):
        super(Bottleneck, self).__init__()
        if norm_layer is None:
            norm_layer = nn.BatchNorm2d
        width = int(planes * (base_width / 64.)) * groups

        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 ResNeXt101_32x4d(nn.Module):

    def __init__(self, num_classes=1000, zero_init_residual=False):
        super(ResNeXt101_32x4d, self).__init__()
        self.num_classes = num_classes
        self.features = resnext101_32x4d_features
        self.avg_pool = nn.AvgPool2d((7, 7), (1, 1))
        self.last_linear = nn.Linear(2048, num_classes)
        self.bottleneck = nn.Sequential(
            nn.Linear(2048, 512),
            nn.BatchNorm1d(512),
            nn.ReLU(),
            nn.Dropout(0.5)
        )
        self.bottleneck[0].weight.data.normal_(0, 0.005)
        self.bottleneck[0].bias.data.fill_(0.1)
        self.head = nn.Sequential(
            nn.Linear(512, 512),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(512, num_classes)
        )
        for dep in range(2):
            self.head[dep * 3].weight.data.normal_(0, 0.01)
            self.head[dep * 3].bias.data.fill_(0.0)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

        if zero_init_residual:
            for m in self.modules():
                if isinstance(m, Bottleneck):
                    nn.init.constant_(m.bn3.weight, 0)
                elif isinstance(m, BasicBlock):
                    nn.init.constant_(m.bn2.weight, 0)

    def logits(self, input):
        # x = self.avg_pool(input)
        # x = x.view(x.size(0), -1)
        x = self.last_linear(input)
        # x = self.bottleneck(input)
        # x = self.head(x)
        return x

    def forward(self, input):
        x = self.features(input)
        x = self.avg_pool(x)
        x = x.view(x.size(0), -1)
        x = self.logits(x)
        return x


class ResNeXt101_64x4d(nn.Module):

    def __init__(self, num_classes=1000, zero_init_residual=False,):
        super(ResNeXt101_64x4d, self).__init__()
        self.num_classes = num_classes
        self.features = resnext101_64x4d_features
        self.avg_pool = nn.AvgPool2d((7, 7), (1, 1))
        # self.last_linear = nn.Linear(2048, num_classes)
        self.out_features = 2048
        self.bottleneck = nn.Sequential(
            nn.Linear(2048, 512),
            nn.BatchNorm1d(512),
            nn.ReLU(),
            nn.Dropout(0.5)
        )
        self.bottleneck[0].weight.data.normal_(0, 0.005)
        self.bottleneck[0].bias.data.fill_(0.1)
        self.head = nn.Sequential(
            nn.Linear(512, 512),
            nn.ReLU(),
            nn.Dropout(0.5),
            nn.Linear(512, num_classes)
        )
        for dep in range(2):
            self.head[dep * 3].weight.data.normal_(0, 0.01)
            self.head[dep * 3].bias.data.fill_(0.0)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

        if zero_init_residual:
            for m in self.modules():
                if isinstance(m, Bottleneck):
                    nn.init.constant_(m.bn3.weight, 0)
                elif isinstance(m, BasicBlock):
                    nn.init.constant_(m.bn2.weight, 0)

    def logits(self, input):
        # x = self.avg_pool(input)
        # x = x.view(x.size(0), -1)
        # x = self.last_linear(x)
        x = self.bottleneck(input)
        x = self.head(x)
        return x


    def forward(self, input):
        x = self.features(input)
        x = self.avg_pool(x)
        x = x.view(x.size(0), -1)  # x为[2xbatchsize,2048]
        """x = self.avg_pool(input)
        x = x.view(x.size(0), -1)
        print("x.shape={}".format(x.shape))
        #x.shape=torch.Size([4, 2048, 7, 7])
        #x = self.logits(x)"""
        return x


def resnext101_32x4d(num_classes=1000, pretrained='imagenet'):
    model = ResNeXt101_32x4d(num_classes=num_classes)
    if pretrained is not None:
        num_classes = 1000
        settings = pretrained_settings['resnext101_32x4d'][pretrained]
        assert num_classes == settings['num_classes'], \
            "num_classes should be {}, but is {}".format(settings['num_classes'], num_classes)
        model.load_state_dict(model_zoo.load_url(settings['url']), strict=False)
        model.input_space = settings['input_space']
        model.input_size = settings['input_size']
        model.input_range = settings['input_range']
        model.mean = settings['mean']
        model.std = settings['std']
    return model


def resnext101_64x4d(num_classes=1000, pretrained='imagenet'):
    model = ResNeXt101_64x4d(num_classes=num_classes)
    if pretrained is not None:
        num_classes = 1000
        settings = pretrained_settings['resnext101_64x4d'][pretrained]
        assert num_classes == settings['num_classes'], \
            "num_classes should be {}, but is {}".format(settings['num_classes'], num_classes)
        model.load_state_dict(model_zoo.load_url(settings['url']), strict=False)
        model.input_space = settings['input_space']
        model.input_size = settings['input_size']
        model.input_range = settings['input_range']
        model.mean = settings['mean']
        model.std = settings['std']
    return model