models.py

import torch.nn as nn
import torch.nn.functional as F
import math


def conv3x3(in_planes, out_planes, stride=1):
    " 3x3 convolution with padding "
    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False)


class BasicBlock(nn.Module):
    expansion=1

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(BasicBlock, self).__init__()
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = nn.BatchNorm2d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = nn.BatchNorm2d(planes)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = 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:
            residual = self.downsample(x)

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

        return out


class BasicBlock_BN(nn.Module):
    expansion=1

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(BasicBlock_BN, self).__init__()
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = nn.BatchNorm2d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = nn.BatchNorm2d(planes)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        x = F.relu(x)
        residual = 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:
            residual = self.downsample(x)

        out += residual

        return out


class ResNet_Cifar(nn.Module):

    def __init__(self, block, layers, num_classes=0):
        super(ResNet_Cifar, self).__init__()
        self.inplanes = 16
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, stride=1, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(16)
        self.relu = nn.ReLU(inplace=True)
        self.layer1 = self._make_layer(block, 16, layers[0])
        self.layer2 = self._make_layer(block, 32, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 64, layers[2], stride=2)
        self.avgpool = nn.AvgPool2d(8, stride=1)
        self.fc = nn.Linear(64 * block.expansion, num_classes)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, math.sqrt(2. / n))
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion)
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for _ in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    def get_bn_before_relu(self):
        if isinstance(self.layer1[0], BasicBlock):
            bn1 = self.layer1[-1].bn2
            bn2 = self.layer2[-1].bn2
            bn3 = self.layer3[-1].bn2
        else:
            print('ResNet_Cifar unknown block error !!!')

        return [bn1, bn2, bn3]

    def forward(self, x, type=''):
        x = self.conv1(x)
        x = self.bn1(x)

        if type in ['AT', 'NST', 'SP', 'FN']:
            x_post = self.relu(x)
            x1 = self.layer1(x_post)
            x2 = self.layer2(x1)
            x3 = self.layer3(x2)

            x = self.avgpool(x3)
            x = x.view(x.size(0), -1)
            x = self.fc(x)

            return x, [x1, x2, x3]
        elif type in ['OD']:
            x1 = self.layer1(x)
            x2 = self.layer2(x1)
            x3 = self.layer3(x2)

            x_post = self.relu(x3)

            x = self.avgpool(x_post)
            x = x.view(x.size(0), -1)
            x = self.fc(x)

            return x, [x1, x2, x3]
        elif type in ['RKD']:
            x_post = self.relu(x)
            x = self.layer1(x_post)
            x = self.layer2(x)
            x = self.layer3(x)

            x = self.avgpool(x)
            x = x.view(x.size(0), -1)
            pen_x = x
            x = self.fc(x)

            return x, pen_x
        else:
            x_post = self.relu(x)
            x = self.layer1(x_post)
            x = self.layer2(x)
            x = self.layer3(x)

            x = self.avgpool(x)
            x = x.view(x.size(0), -1)
            x = self.fc(x)

            return x


class ConvReg(nn.Module):
    """Convolutional regression for FitNet"""
    def __init__(self, s_shape, t_shape, use_relu=True):
        super(ConvReg, self).__init__()
        self.use_relu = use_relu
        s_N, s_C, s_H, s_W = s_shape
        t_N, t_C, t_H, t_W = t_shape
        if s_H == 2 * t_H:
            self.conv = nn.Conv2d(s_C, t_C, kernel_size=3, stride=2, padding=1)
        elif s_H * 2 == t_H:
            self.conv = nn.ConvTranspose2d(s_C, t_C, kernel_size=4, stride=2, padding=1)
        elif s_H >= t_H:
            self.conv = nn.Conv2d(s_C, t_C, kernel_size=(1+s_H-t_H, 1+s_W-t_W))
        else:
            raise NotImplemented('student size {}, teacher size {}'.format(s_H, t_H))
        self.bn = nn.BatchNorm2d(t_C)
        self.relu = nn.ReLU(inplace=True)

    def forward(self, x):
        x = self.conv(x)
        if self.use_relu:
            return self.relu(self.bn(x))
        else:
            return self.bn(x)