`nexuml_library.layers.model.resnet`¶

nexuml_library.layers.model.resnet ¶

ResNet pipeline layer for image classification backbones.

ResNet backbone that emits feature embeddings.

Wraps torchvision.models.resnet* and replaces the final fully-connected layer with an identity so downstream heads can operate on embeddings.

Attributes:

Name	Type	Description
`resnet_type`	`str`	Architecture variant (e.g. `"resnet18"`).
`pretrained`	`bool`	Whether to load pretrained ImageNet weights.
`cifar_stem`	`bool \| None`	Use a CIFAR-friendly stem (3x3 stride-1, no maxpool). Defaults to `False` when pretrained, `True` otherwise.