piscis.networks.fpn

piscis.networks.fpn#

Attributes#

`ModuleDef`
`BatchActConv`
`BatchConv`

Classes#

`BatchConvStyle`	Convolutional block with batch norm, activation, and style transfer.
`UpConv`	Upsampling convolutional block.
`MakeStyle`	Style transfer module.
`Decoder`	Decoder module.
`FPN`	Feature pyramid network.

Module Contents#

piscis.networks.fpn.ModuleDef#

piscis.networks.fpn.BatchActConv#

piscis.networks.fpn.BatchConv#

class piscis.networks.fpn.BatchConvStyle(in_channels: int, out_channels: int, style_channels: int, kernel_size: int | Sequence[int], conv: ModuleDef, dense: ModuleDef, bn: ModuleDef, act: ModuleDef)#

Bases: torch.nn.Module

Convolutional block with batch norm, activation, and style transfer.

Parameters:

in_channelsint: Number of input channels.
out_channelsint: Number of output channels.
style_channelsint: Number of style channels.
kernel_sizeUnion[int, Sequence[int]]: Size of the convolutional kernel.
convModuleDef: Convolution module.
denseModuleDef: Dense module.
bnModuleDef: Batch norm module.
actModuleDef: Activation function.

conv#

dense#

forward(style: torch.Tensor | None, x: torch.Tensor, y: torch.Tensor | None = None) → torch.Tensor#

class piscis.networks.fpn.UpConv(in_channels: int, out_channels: int, style_channels: int, kernel_size: int | Sequence[int], conv: ModuleDef, dense: ModuleDef, bn: ModuleDef, act: ModuleDef)#

Bases: torch.nn.Module

Upsampling convolutional block.

Parameters:

in_channelsint: Number of input channels.
out_channelsint: Number of output channels.
style_channelsint: Number of style channels.
kernel_sizeUnion[int, Sequence[int]]: Size of the convolutional kernel.
convModuleDef: Convolution module.
denseModuleDef: Dense module.
bnModuleDef: Batch norm module.
actModuleDef: Activation function.

proj#

conv#

convs_0#

convs_1#

convs_2#

forward(x: torch.Tensor, y: torch.Tensor | None, style: torch.Tensor | None) → torch.Tensor#

class piscis.networks.fpn.MakeStyle(*args, **kwargs)#

Bases: torch.nn.Module

Style transfer module.

forward(x: torch.Tensor) → torch.Tensor#

class piscis.networks.fpn.Decoder(stage_sizes: Sequence[int], kernel_size: int | Sequence[int], conv: ModuleDef, dense: ModuleDef, bn: ModuleDef, act: ModuleDef)#

Bases: torch.nn.Module

Decoder module.

Parameters:

stage_sizesSequence[int]: Number of channels at each stage.
kernel_sizeUnion[int, Sequence[int]]: Size of the convolutional kernel.
convModuleDef: Convolution module.
denseModuleDef: Dense module.
bnModuleDef: Batch norm module.
actModuleDef: Activation function.

stage_sizes#

upsample#

up_blocks#

resize_up_blocks#

out_channels#

forward(style: torch.Tensor, xd: Sequence[torch.Tensor]) → torch.Tensor#

class piscis.networks.fpn.FPN(encoder: ModuleDef, encoder_levels: Sequence[int], in_channels: int, out_channels: int, kernel_size: int | Sequence[int] = 3, style: bool = True, bn_momentum: float = 0.1, bn_epsilon: float = 1e-05, conv: ModuleDef = nn.Conv2d, dense: ModuleDef = nn.Linear, bn: ModuleDef = nn.BatchNorm2d, act: ModuleDef = nn.SiLU)#

Bases: torch.nn.Module

Feature pyramid network.

Parameters:

encoderModuleDef: Encoder module.
encoder_levelsSequence[int]: Encoder levels to use for the feature pyramid.
in_channelsint: Number of input channels.
out_channelsint: Number of output channels.
kernel_sizeUnion[int, Sequence[int]], optional: Size of the convolutional kernel. Default is s3.
stylebool, optional: Whether to use style transfer. Default is True.
bn_momentumfloat, optional: Momentum parameter for batch norm layers. Default is 0.1.
bn_epsilonfloat, optional: Epsilon parameter for batch norm layers. Default is 1e-5.
convModuleDef, optional: Convolution module. Default is nn.Conv2d.
denseModuleDef, optional: Dense module. Default is nn.Linear.
bnModuleDef, optional: Batch norm module. Default is nn.BatchNorm2d.
actModuleDef, optional: Activation function. Default is nn.SiLU.

encoder#

encoder_levels#

style = True#

decoder#

output#

forward(x: torch.Tensor) → Tuple[torch.Tensor, torch.Tensor | None]#