Pretrained weights for lucid.models.mask2former_swin_base.
Single ADE20k semantic checkpoint converted from
facebook/mask2former-swin-base-ade-semantic (Cheng et al., 2022;
~107M params, 52.4 mIoU on the ADE20k validation set).
Examples
>>> from lucid.models.vision.mask2former import Mask2FormerSwinBaseWeights
>>> list(Mask2FormerSwinBaseWeights.__members__)
['ADE20K', 'DEFAULT']
DEFAULT is an alias for the tag a bare
pretrained=True resolves to, so the two ways of asking for
weights cannot drift apart.