Pretrained weights for lucid.models.fcn_resnet101.
Single COCO-with-VOC-labels checkpoint converted from reference_vision's
FCN_ResNet101_Weights.COCO_WITH_VOC_LABELS_V1 (Long et al., 2015;
~54.3M params, 63.7 mIoU / 91.9 pixel acc).
Examples
>>> from lucid.models.vision.fcn import FCNResNet101Weights
>>> list(FCNResNet101Weights.__members__)
['COCO_WITH_VOC_LABELS_V1', '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.