SegFormerBaseline

class torch_uncertainty.baselines.segmentation.SegFormerBaseline(num_classes, loss, version, arch, num_estimators=1)

SegFormer backbone baseline for segmentation providing support for various versions and architectures.

Parameters:

• num_classes (int) – Number of classes to predict.

• loss (type[Module]) – Training loss.

• version (str) –

  Determines which SegFormer version to use. Options are:

  • "std": original SegFormer

• arch (int) –

  Determines which architecture to use. Options are:

  • 0: SegFormer-B0

  • 1: SegFormer-B1

  • 2: SegFormer-B2

  • 3: SegFormer-B3

  • 4: SegFormer-B4

  • 5: SegFormer-B5

• num_estimators (int, optional) – Number of estimators in the ensemble. Defaults to 1 (single model).