def __init__(self, in_channels, out_channels, kernel_size=3, activation=None, upsampling=1):
super().__init__()
self.s1 = nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, padding=kernel_size // 2),
nn.UpsamplingBilinear2d(scale_factor=upsampling) if upsampling > 1 else nn.Identity(),
Activation(activation))
self.s2 = nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, padding=kernel_size // 2),
nn.UpsamplingBilinear2d(scale_factor=upsampling) if upsampling > 1 else nn.Identity(),
Activation(activation))
self.s3 = nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, padding=kernel_size // 2),
nn.UpsamplingBilinear2d(scale_factor=upsampling) if upsampling > 1 else nn.Identity(),
Activation(activation))
self.s4 = nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, padding=kernel_size // 2),
nn.UpsamplingBilinear2d(scale_factor=upsampling) if upsampling > 1 else nn.Identity(),
Activation(activation))
def __init__(
self,
encoder_weights: str = None,
encoder_freeze: bool = True,
decoder_use_batchnorm: bool = True,
decoder_channels: List[int] = (256, 128, 64, 32, 16),
classes: int = 1,
activation: Optional[Union[str, callable]] = "sigmoid",
device=torch.device("cpu"),
):
super().__init__()
self.encoder = moco_r50(encoder_weights,
encoder_freeze=encoder_freeze,
map_location=device)
self.decoder = UnetDecoder(
encoder_channels=(12, 64, 256, 512, 1024, 2048),
decoder_channels=decoder_channels,
n_blocks=5,
use_batchnorm=decoder_use_batchnorm,
center=False,
attention_type=None,
)
self.segmentation_head = torch.nn.Sequential(
torch.nn.AdaptiveAvgPool2d((1, 1)),
torch.nn.Flatten(),
torch.nn.Linear(16, classes),
Activation(activation),
)
self.classification_head = None
self.initialize()
def __init__(self, in_channels, out_channels, kernel_size=3):
conv2d = nn.Conv2d(in_channels,
out_channels,
kernel_size,
padding=kernel_size // 2)
identity = nn.Identity()
activation = Activation(None)
super().__init__(conv2d, identity, activation)
def __init__(self,
eps=1.,
activation=None,
ignore_channels=None,
**kwargs):
super().__init__(**kwargs)
self.eps = eps
self.activation = Activation(activation)
self.ignore_channels = ignore_channels
def __init__(self,
eps=1e-7,
threshold=0.5,
activation=None,
ignore_channels=None,
**kwargs):
super().__init__(**kwargs)
self.eps = eps
self.threshold = threshold
self.activation = Activation(activation)
self.ignore_channels = ignore_channels
def __init__(self,
eps=1.,
alf=0.3,
beta=0.7,
activation=None,
ignore_channels=None,
**kwargs):
super().__init__(**kwargs)
self.eps = eps
self.alf = alf
self.beta = beta
self.activation = Activation(activation)
self.ignore_channels = ignore_channels
def __init__(self, bce=1.0, dice=0.5, lovasz=0.0, soft_bce=True):
super().__init__()
if soft_bce:
self.bce = losses.soft_bce.SoftBCEWithLogitsLoss(smooth_factor=0.1)
else:
self.bce = nn.BCEWithLogitsLoss()
self.dice = DiceLoss(activation=None)
self.lovasz = losses.lovasz.LovaszLoss(mode='binary',
from_logits=False)
self.activation = Activation('sigmoid')
self.w_bce = bce
self.w_dice = dice
self.w_lovasz = lovasz
def __init__(self,
eps=1.,
beta=1.,
weight=1,
activation=None,
ignore_channels=None,
**kwargs):
super().__init__(**kwargs)
self.eps = eps
self.beta = beta
self.activation = Activation(activation)
self.ignore_channels = ignore_channels
self.weight = weight
def __init__(self,
in_channels,
out_channels,
kernel_size=3,
activation=None,
upsampling=1):
dropout = nn.Dropout2d(0.5)
conv2d = nn.Conv2d(in_channels,
out_channels,
kernel_size=kernel_size,
padding=kernel_size // 2)
upsampling = nn.UpsamplingBilinear2d(
scale_factor=upsampling) if upsampling > 1 else nn.Identity()
activation = Activation(activation)
super().__init__(dropout, conv2d, upsampling, activation)