154 lines
5.3 KiB
Python
154 lines
5.3 KiB
Python
# Resnet implementation that adds a u-net style up-conversion component to output values at a
|
|
# specified pixel density.
|
|
#
|
|
# The downsampling part of the network is compatible with the built-in torch resnet for use in
|
|
# transfer learning.
|
|
#
|
|
# Only resnet50 currently supported.
|
|
|
|
import torch
|
|
import torch.nn as nn
|
|
from torchvision.models.resnet import BasicBlock, Bottleneck, conv1x1, conv3x3
|
|
from torchvision.models.utils import load_state_dict_from_url
|
|
import torchvision
|
|
|
|
|
|
from trainer.networks import register_model
|
|
from utils.util import checkpoint
|
|
|
|
model_urls = {
|
|
'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
|
|
}
|
|
|
|
|
|
class ReverseBottleneck(nn.Module):
|
|
|
|
def __init__(self, inplanes, planes, groups=1, passthrough=False,
|
|
base_width=64, dilation=1, norm_layer=None):
|
|
super().__init__()
|
|
if norm_layer is None:
|
|
norm_layer = nn.BatchNorm2d
|
|
width = int(planes * (base_width / 64.)) * groups
|
|
self.passthrough = passthrough
|
|
if passthrough:
|
|
self.integrate = conv1x1(inplanes*2, inplanes)
|
|
self.bn_integrate = norm_layer(inplanes)
|
|
# Both self.conv2 and self.downsample layers downsample the input when stride != 1
|
|
self.conv1 = conv1x1(inplanes, width)
|
|
self.bn1 = norm_layer(width)
|
|
self.conv2 = conv3x3(width, width, groups, dilation)
|
|
self.bn2 = norm_layer(width)
|
|
self.residual_upsample = nn.Sequential(
|
|
nn.Upsample(scale_factor=2, mode='nearest'),
|
|
conv1x1(width, width),
|
|
norm_layer(width),
|
|
)
|
|
self.conv3 = conv1x1(width, planes)
|
|
self.bn3 = norm_layer(planes)
|
|
self.relu = nn.ReLU(inplace=True)
|
|
self.upsample = nn.Sequential(
|
|
nn.Upsample(scale_factor=2, mode='nearest'),
|
|
conv1x1(inplanes, planes),
|
|
norm_layer(planes),
|
|
)
|
|
|
|
def forward(self, x, passthrough=None):
|
|
if self.passthrough:
|
|
x = self.bn_integrate(self.integrate(torch.cat([x, passthrough], dim=1)))
|
|
|
|
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.residual_upsample(out)
|
|
|
|
out = self.conv3(out)
|
|
out = self.bn3(out)
|
|
|
|
identity = self.upsample(x)
|
|
|
|
out = out + identity
|
|
out = self.relu(out)
|
|
|
|
return out
|
|
|
|
|
|
class UResNet50(torchvision.models.resnet.ResNet):
|
|
|
|
def __init__(self, block, layers, num_classes=1000, zero_init_residual=False,
|
|
groups=1, width_per_group=64, replace_stride_with_dilation=None,
|
|
norm_layer=None):
|
|
super().__init__(block, layers, num_classes, zero_init_residual, groups, width_per_group,
|
|
replace_stride_with_dilation, norm_layer)
|
|
if norm_layer is None:
|
|
norm_layer = nn.BatchNorm2d
|
|
'''
|
|
# For reference:
|
|
self.layer1 = self._make_layer(block, 64, layers[0])
|
|
self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
|
|
dilate=replace_stride_with_dilation[0])
|
|
self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
|
|
dilate=replace_stride_with_dilation[1])
|
|
self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
|
|
dilate=replace_stride_with_dilation[2])
|
|
'''
|
|
uplayers = []
|
|
inplanes = 2048
|
|
first = True
|
|
for i in range(2):
|
|
uplayers.append(ReverseBottleneck(inplanes, inplanes // 2, norm_layer=norm_layer, passthrough=not first))
|
|
inplanes = inplanes // 2
|
|
first = False
|
|
self.uplayers = nn.ModuleList(uplayers)
|
|
self.tail = nn.Sequential(conv1x1(1024, 512),
|
|
norm_layer(512),
|
|
nn.ReLU(),
|
|
conv3x3(512, 512),
|
|
norm_layer(512),
|
|
nn.ReLU(),
|
|
conv1x1(512, 128))
|
|
|
|
del self.fc # Not used in this implementation and just consumes a ton of GPU memory.
|
|
|
|
|
|
def _forward_impl(self, x):
|
|
# Should be the exact same implementation of torchvision.models.resnet.ResNet.forward_impl,
|
|
# except using checkpoints on the body conv layers.
|
|
x = self.conv1(x)
|
|
x = self.bn1(x)
|
|
x = self.relu(x)
|
|
x = self.maxpool(x)
|
|
|
|
x1 = checkpoint(self.layer1, x)
|
|
x2 = checkpoint(self.layer2, x1)
|
|
x3 = checkpoint(self.layer3, x2)
|
|
x4 = checkpoint(self.layer4, x3)
|
|
unused = self.avgpool(x4) # This is performed for instance-level pixpro learning, even though it is unused.
|
|
|
|
x = checkpoint(self.uplayers[0], x4)
|
|
x = checkpoint(self.uplayers[1], x, x3)
|
|
#x = checkpoint(self.uplayers[2], x, x2)
|
|
#x = checkpoint(self.uplayers[3], x, x1)
|
|
|
|
return checkpoint(self.tail, torch.cat([x, x2], dim=1))
|
|
|
|
def forward(self, x):
|
|
return self._forward_impl(x)
|
|
|
|
|
|
@register_model
|
|
def register_u_resnet50(opt_net, opt):
|
|
model = UResNet50(Bottleneck, [3, 4, 6, 3])
|
|
return model
|
|
|
|
|
|
if __name__ == '__main__':
|
|
model = UResNet50(Bottleneck, [3,4,6,3])
|
|
samp = torch.rand(1,3,224,224)
|
|
model(samp)
|
|
# For pixpro: attach to "tail.3"
|