3D fcnn

[tommycwong]

Updated fcnn blocks with 3D versions. I am wondering if separate 3D versions of fcnns should be created or if existing fcnns should be updated with ndim arg in fcnn.py.

[ziatdinovmax]

Thank you, Tommy. I think 3D convolutions can be simply used as drop-ins. Can you please write it in a more compact way, without if/else? For example,

def get_conv(dim: int) -> Type[nn.Module]:
    conv_dict = {1: nn.Conv1d, 2: nn.Conv2d, 3: nn.Conv3d}
    return conv_dict[dim]

PS. Not sure whether List is a correct type annotation for **kwargs.

[tommycwong]

I am not sure what output type I should put for the type annotation.

[ziatdinovmax]

@tommycwong I think you need to define get_conv and get_BatchNorm (and also MaxPool?) only once, outside of all classes, and then just call those functions from each class.

[ziatdinovmax]

I am not sure what output type I should put for the type annotation.

nn.Module