import math
import torch
import torch.nn as nn
import torch.nn.functional as F
__all__ = ['resnet152_fd', ]
def tf_pad(x, kernel_size=3, stride=2):
"""For stride = 2 or stride = 3"""
if x.shape[2] % stride == 0:
padH = max(kernel_size - stride, 0)
else:
padH = max(kernel_size - (x.shape[2] % stride), 0)
if x.shape[3] % stride == 0:
padW = max(kernel_size - stride, 0)
else:
padW = max(kernel_size - (x.shape[3] % stride), 0)
pad_top = padH // 2
pad_bottom = padH - pad_top
pad_left = padW // 2
pad_right = padW - pad_left
x_padded = torch.nn.functional.pad(
x, pad=(pad_left, pad_right, pad_top, pad_bottom))
return x_padded
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=0, 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 NonLocal(nn.Module):
def __init__(self, channels, inplace=False, embed=True, softmax=True):
super().__init__()
self.channels = channels
self.inplace = inplace
self.embed = embed
self.softmax = softmax
if self.embed:
self.embedding_theta = nn.Conv2d(channels, channels//2, kernel_size=1, stride=1, bias=False)
self.embedding_phi = nn.Conv2d(channels, channels//2, kernel_size=1, stride=1, bias=False)
self.conv = nn.Conv2d(channels, channels, kernel_size=1, stride=1, bias=False)
self.bn = nn.BatchNorm2d(channels)
def forward(self, x):
n, c, h, w = x.shape
res = x
if self.embed:
theta = self.embedding_theta(x)
phi = self.embedding_phi(x)
g = x
else:
theta, phi, g = x, x, x
if c > h * w or self.softmax:
f = torch.einsum('niab,nicd->nabcd', theta, phi)
if self.softmax:
orig_shape = f.shape
f = f.reshape(-1, h * w, h * w)
f = f / torch.sqrt(torch.tensor(c, device=f.device, dtype=f.dtype))
f = F.softmax(f, dim=-1)
f = f.reshape(orig_shape)
f = torch.einsum('nabcd,nicd->niab', f, g)
else:
f = torch.einsum('nihw,njhw->nij', phi, g)
f = torch.einsum('nij,nihw->njhw', f, theta)
if not self.softmax:
f = f / torch.tensor(h * w, device=f.device, dtype=f.dtype)
f = f.reshape(x.shape)
f = self.conv(f)
f = self.bn(f)
return f + res
class Bottleneck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, **kwargs):
super(Bottleneck, self).__init__()
self.conv1 = conv1x1(inplanes, planes,)
self.bn1 = nn.BatchNorm2d(planes)
self.conv2 = conv3x3(planes, planes, stride=stride)
self.bn2 = nn.BatchNorm2d(planes)
self.conv3 = conv1x1(planes, planes * 4)
self.bn3 = nn.BatchNorm2d(planes * 4)
self.relu = nn.GELU()
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 = tf_pad(out, kernel_size=3, stride=self.stride)
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:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
class ResNetDenoiseModel(nn.Module):
def __init__(self, block, layers, num_classes=1000, denoise=True, **kwargs):
super(ResNetDenoiseModel, self).__init__()
self.inplanes = 64
self.denoise = denoise
self.conv0 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=0, bias=False)
self.bn0 = nn.BatchNorm2d(64)
self.relu = nn.GELU()
self.pool0 = nn.MaxPool2d(kernel_size=3, stride=2, padding=0)
self.group0 = self._make_layer(block, 64, layers[0], stride=1)
self.group1 = self._make_layer(block, 128, layers[1], stride=2)
self.group2 = self._make_layer(block, 256, layers[2], stride=2)
self.group3 = self._make_layer(block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.linear = nn.Linear(512 * block.expansion, num_classes, bias=True)
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(
conv1x1(self.inplanes, planes * block.expansion, stride),
nn.BatchNorm2d(planes * block.expansion),
)
layers = [block(self.inplanes, planes, stride, downsample)]
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(self.inplanes, planes))
if self.denoise:
layers.append(NonLocal(planes * 4))
return nn.Sequential(*layers)
def forward(self, x):
x = x[:, [2,1,0], :, :]
x = tf_pad(x, kernel_size=7, stride=2)
x = self.conv0(x)
x = self.bn0(x)
x = self.relu(x)
x = tf_pad(x, kernel_size=3, stride=2)
x = self.pool0(x)
x = self.group0(x)
x = self.group1(x)
x = self.group2(x)
x = self.group3(x)
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.linear(x)
return x
def _resnet(arch, block, layers, **kwargs):
model = ResNetDenoiseModel(block, layers, **kwargs)
return model
[docs]def resnet152_fd(**kwargs):
'''The function to create resnet152 model of feature denoising.'''
kwargs['denoise'] = True
return _resnet('resnet152_fd', Bottleneck, [3, 8, 36, 3], **kwargs)