Networks

End-to-end classification and general-purpose networks composing the modules above, plus the UNet-ConvNeXt baselines used in benchmark comparisons.

Classification & general-purpose

ClassificationResNet(in_channels, ...[, ...])	Residual network with global-average-pool readout for classification.
ResidualNetwork(in_channels, out_channels, ...)	General-purpose residual network backbone (see module docstring for architecture).
ViT5ClassificationNet(in_channels, ...[, ...])	ViT-5 classification network.

Baselines

UNet-ConvNeXt baselines ported from The Well, used as PDE-task reference points. UNetConvNextV2 fixes the upstream finest-skip bug; see the module docstring for the diff.

UNetConvNext(dim_in, dim_out, n_spatial_dims)	UNet with ConvNeXt blocks — channels-first (NCHW / NCDHW) interface.
WellUNetConvNext(**kwargs)	UNet-ConvNeXt with the dict-based channels-last interface expected by WELLRegressionWrapper.
UNetConvNextV2(dim_in, dim_out, n_spatial_dims)	UNet-ConvNeXt with corrected skip connections.
WellUNetConvNextV2(**kwargs)	Like WellUNetConvNext but with fixed skip connections.