I have "simple" Unet with resnet encoder on pytorch v3.1, which works pretty fine:
class UNetResNet(nn.Module):
"""PyTorch U-Net model using ResNet(34, 101 or 152) encoder.
UNet: https://arxiv.org/abs/1505.04597
ResNet: https://arxiv.org/abs/1512.03385
Args:
encoder_depth (int): Depth of a ResNet encoder (34, 101 or 152).
num_classes (int): Number of output classes.
num_filters (int, optional): Number of filters in the last layer of decoder. Defaults to 32.
dropout_2d (float, optional): Probability factor of dropout layer before output layer. Defaults to 0.2.
pretrained (bool, optional):
False - no pre-trained weights are being used.
True - ResNet encoder is pre-trained on ImageNet.
Defaults to False.
is_deconv (bool, optional):
False: bilinear interpolation is used in decoder.
True: deconvolution is used in decoder.
Defaults to False.
"""
def __init__(self, encoder_depth, num_classes, num_filters=32, dropout_2d=0.2,
pretrained=False, is_deconv=False):
super().__init__()
self.num_classes = num_classes
self.dropout_2d = dropout_2d
if encoder_depth == 34:
self.encoder = torchvision.models.resnet34(pretrained=pretrained)
bottom_channel_nr = 512
elif encoder_depth == 101:
self.encoder = torchvision.models.resnet101(pretrained=pretrained)
bottom_channel_nr = 2048
elif encoder_depth == 152:
self.encoder = torchvision.models.resnet152(pretrained=pretrained)
bottom_channel_nr = 2048
else:
raise NotImplementedError('only 34, 101, 152 version of Resnet are implemented')
self.pool = nn.MaxPool2d(2, 2)
self.relu = nn.ReLU(inplace=True)
self.conv1 = nn.Sequential(self.encoder.conv1,
self.encoder.bn1,
self.encoder.relu,
self.pool)
self.conv2 = self.encoder.layer1
self.conv3 = self.encoder.layer2
self.conv4 = self.encoder.layer3
self.conv5 = self.encoder.layer4
self.center = DecoderBlockV2(bottom_channel_nr, num_filters * 8 * 2, num_filters * 8, is_deconv)
self.dec5 = DecoderBlockV2(bottom_channel_nr + num_filters * 8, num_filters * 8 * 2, num_filters * 8, is_deconv)
self.dec4 = DecoderBlockV2(bottom_channel_nr // 2 + num_filters * 8, num_filters * 8 * 2, num_filters * 8,
is_deconv)
self.dec3 = DecoderBlockV2(bottom_channel_nr // 4 + num_filters * 8, num_filters * 4 * 2, num_filters * 2,
is_deconv)
self.dec2 = DecoderBlockV2(bottom_channel_nr // 8 + num_filters * 2, num_filters * 2 * 2, num_filters * 2 * 2,
is_deconv)
self.dec1 = DecoderBlockV2(num_filters * 2 * 2, num_filters * 2 * 2, num_filters, is_deconv)
self.dec0 = ConvRelu(num_filters, num_filters)
self.final = nn.Conv2d(num_filters, num_classes, kernel_size=1)
def forward(self, x):
conv1 = self.conv1(x)
conv2 = self.conv2(conv1)
conv3 = self.conv3(conv2)
conv4 = self.conv4(conv3)
conv5 = self.conv5(conv4)
pool = self.pool(conv5) # that pool layer I would like to delete
center = self.center(pool)
dec5 = self.dec5(torch.cat([center, conv5], 1))
dec4 = self.dec4(torch.cat([dec5, conv4], 1))
dec3 = self.dec3(torch.cat([dec4, conv3], 1))
dec2 = self.dec2(torch.cat([dec3, conv2], 1))
dec1 = self.dec1(dec2)
dec0 = self.dec0(dec1)
return self.final(F.dropout2d(dec0, p=self.dropout_2d))
By deleting that pool layer from forward we come to next forward function:
def forward(self, x):
conv1 = self.conv1(x)
conv2 = self.conv2(conv1)
conv3 = self.conv3(conv2)
conv4 = self.conv4(conv3)
conv5 = self.conv5(conv4)
center = self.center(conv5) #now center connects with conv5
dec5 = self.dec5(torch.cat([center, conv5], 1))
dec4 = self.dec4(torch.cat([dec5, conv4], 1))
dec3 = self.dec3(torch.cat([dec4, conv3], 1))
dec2 = self.dec2(torch.cat([dec3, conv2], 1))
dec1 = self.dec1(dec2)
dec0 = self.dec0(dec1)
return self.final(F.dropout2d(dec0, p=self.dropout_2d))
Applying only such changes and we will face such error:
~/Desktop/ml/salt/unet_models.py in forward(self, x)
397 center = self.center(conv5)
398
--> 399 dec5 = self.dec5(torch.cat([center, conv5], 1))
400
401 dec4 = self.dec4(torch.cat([dec5, conv4], 1))
RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1. Got 4 and 8 in dimension 2 at /pytorch/torch/lib/THC/generic/THCTensorMath.cu:111
So we have to somehow change dec4, but how?
More info:
class DecoderBlockV2(nn.Module):
def __init__(self, in_channels, middle_channels, out_channels, is_deconv=True):
super(DecoderBlockV2, self).__init__()
self.in_channels = in_channels
if is_deconv:
"""
Paramaters for Deconvolution were chosen to avoid artifacts, following
link https://distill.pub/2016/deconv-checkerboard/
"""
self.block = nn.Sequential(
ConvRelu(in_channels, middle_channels),
nn.ConvTranspose2d(middle_channels, out_channels, kernel_size=4, stride=2,
padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True)
)
else:
self.block = nn.Sequential(
nn.Upsample(scale_factor=2, mode='bilinear'),
ConvRelu(in_channels, middle_channels),
ConvRelu(middle_channels, out_channels),
)
def forward(self, x):
return self.block(x)
class ConvRelu(nn.Module):
def __init__(self, in_, out):
super().__init__()
self.conv = conv3x3(in_, out)
self.activation = nn.ReLU(inplace=True)
def forward(self, x):
x = self.conv(x)
x = self.activation(x)
return x