mask2former_swin_tiny(pretrained: bool | str = False, weights: Mask2FormerSwinTinyWeights | None = None, overrides: object = {})Mask2Former with Swin-Tiny backbone (Cheng et al., CVPR 2022).
Builds a Mask2FormerForSemanticSegmentation with the Swin-Tiny
backbone (embed_dim = 96, depths = (2, 2, 6, 2),
num_heads = (3, 6, 12, 24), window_size = 7), a 6-layer
multi-scale deformable pixel decoder, and a 9-layer masked-attention
transformer decoder with 100 queries. Default targets ADE20K
(150 classes); reaches ADE20K validation mIoU of 47.7%.
Model Size
Parameters
pretrainedbool or str= FalsePretrained-weight selector.
False → random init; True
→ the DEFAULT tag (Mask2FormerSwinTinyWeights.ADE20K);
a tag string → that specific checkpoint. Mutually exclusive with
weights (which wins if both are given).Explicit weights enum member. Takes precedence over
pretrained.**overridesobject= {}Keyword overrides forwarded into
Mask2FormerConfig.Returns
Mask2FormerForSemanticSegmentationSegmentation model with the Mask2Former-Swin-T configuration applied
(or with overrides merged on top of it).
Examples
>>> import lucid
>>> from lucid.models.vision.mask2former import mask2former_swin_tiny
>>> model = mask2former_swin_tiny()
>>> x = lucid.randn(1, 3, 384, 384)
>>> out = model(x)
>>> out.logits.shape[1]
150