MR/Model.py

from konfai.network import network
import segmentation_models_pytorch as smp
import torch

class Head(network.ModuleArgsDict):

    def __init__(self):
        super().__init__()
        self.add_module("Tanh", torch.nn.Tanh())

class UNetpp(network.Network):

    # 5 input channels = a 2.5D stack of 5 adjacent slices. This is intrinsic to the pretrained weights,
    # not a tunable — so it is fixed here and never exposed in the config.
    IN_CHANNELS = 5

    def __init__(self,
                optimizer : network.OptimizerLoader = network.OptimizerLoader(),
                schedulers: dict[str, network.LRSchedulersLoader] = {
                    "default:ReduceLROnPlateau": network.LRSchedulersLoader(0)
                },
                outputs_criterions: dict[str, network.TargetCriterionsLoader] = {"default" : network.TargetCriterionsLoader()}):
        super().__init__(in_channels = UNetpp.IN_CHANNELS, optimizer = optimizer, schedulers = schedulers, outputs_criterions = outputs_criterions, dim = 2)
        self.add_module("model", smp.UnetPlusPlus(
            encoder_name="resnet34",
            encoder_weights=None,
            in_channels=UNetpp.IN_CHANNELS,
            classes=1,  
            activation=None  
        ))
        self.add_module("Head", Head())