"""Adapted from DPImageBench: https://github.com/fjxmlzn/DPImageBench/blob/main/evaluation/classifier/wrn.py"""
# import math
import torch
import torch.nn as nn
import torch.nn.functional as F
[docs]
class BasicBlock(nn.Module):
def __init__(self, in_planes, out_planes, stride, dropRate=0.0):
super(BasicBlock, self).__init__()
self.bn1 = nn.BatchNorm2d(in_planes)
self.relu1 = nn.ReLU(inplace=True)
self.conv1 = nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(out_planes)
self.relu2 = nn.ReLU(inplace=True)
self.conv2 = nn.Conv2d(out_planes, out_planes, kernel_size=3, stride=1, padding=1, bias=False)
self.droprate = dropRate
self.equalInOut = in_planes == out_planes
self.convShortcut = (
(not self.equalInOut)
and nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, padding=0, bias=False)
or None
)
[docs]
def forward(self, x):
if not self.equalInOut:
x = self.relu1(self.bn1(x))
else:
out = self.relu1(self.bn1(x))
out = self.relu2(self.bn2(self.conv1(out if self.equalInOut else x)))
if self.droprate > 0:
out = F.dropout(out, p=self.droprate, training=self.training)
out = self.conv2(out)
return torch.add(x if self.equalInOut else self.convShortcut(x), out)
[docs]
class NetworkBlock(nn.Module):
def __init__(self, nb_layers, in_planes, out_planes, block, stride, dropRate=0.0):
super(NetworkBlock, self).__init__()
self.layer = self._make_layer(block, in_planes, out_planes, nb_layers, stride, dropRate)
def _make_layer(self, block, in_planes, out_planes, nb_layers, stride, dropRate):
layers = []
for i in range(int(nb_layers)):
layers.append(block(i == 0 and in_planes or out_planes, out_planes, i == 0 and stride or 1, dropRate))
return nn.Sequential(*layers)
[docs]
def forward(self, x):
return self.layer(x)
[docs]
class WideResNet(nn.Module):
def __init__(self, in_c=3, img_size=32, num_classes=10, depth=40, widen_factor=4, dropRate=0.0):
super(WideResNet, self).__init__()
nChannels = [16, 16 * widen_factor, 32 * widen_factor, 64 * widen_factor]
assert (depth - 4) % 6 == 0
n = (depth - 4) / 6
block = BasicBlock
self.img_size = img_size
# 1st conv before any network block
self.conv1 = nn.Conv2d(in_c, nChannels[0], kernel_size=3, stride=1, padding=1, bias=False)
# 1st block
self.block1 = NetworkBlock(n, nChannels[0], nChannels[1], block, 1, dropRate)
# 2nd block
self.block2 = NetworkBlock(n, nChannels[1], nChannels[2], block, 2, dropRate)
# 3rd block
self.block3 = NetworkBlock(n, nChannels[2], nChannels[3], block, 2, dropRate)
# global average pooling and classifier
self.bn1 = nn.BatchNorm2d(nChannels[3])
self.relu = nn.ReLU(inplace=True)
self.fc = nn.Linear(nChannels[3], num_classes)
self.nChannels = nChannels[3]
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):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.bias.data.zero_()
[docs]
def forward(self, x):
out = self.conv1(x)
out = self.block1(out)
out = self.block2(out)
out = self.block3(out)
out = self.relu(self.bn1(out))
out = F.avg_pool2d(out, self.img_size // 4)
out = out.view(-1, self.nChannels)
return self.fc(out)
if __name__ == "__main__":
model = WideResNet(in_c=1, img_size=28)
images = torch.randn((4, 1, 28, 28))
out = model(images)
print(out.shape)