Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels

Benchmark Model Rank Results
semi-supervised-semantic-segmentation-on-pascal-voc-2012-50U2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, CutMix)#3Validation mIoU: 80.5%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-12-5-labeledU2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, CutMix)#10Validation mIoU: 79.01%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-25-labeledU2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, CutMix)#10Validation mIoU: 79.3
semi-supervised-semantic-segmentation-on-pascal-voc-2012-6-25-labeledU2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, CutMix)#10Validation mIoU: 77.21
semi-supervised-semantic-segmentation-on-cityscapes-25-labeledU2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, AEL)#12Validation mIoU: 78.51%
semi-supervised-semantic-segmentation-on-cityscapes-6-25-labeledU2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, AEL)#12Validation mIoU: 74.90%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-366-labeledU2PL (DeepLab v3+ with ResNet-101)#12Validation mIoU: 73.7
semi-supervised-semantic-segmentation-on-cityscapes-50-labeledU2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, AEL)#13Validation mIoU: 79.12%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-1464-labelsU2PL (DeepLab v3+ with ResNet-101)#13Validation mIoU: 79.5
semi-supervised-semantic-segmentation-on-pascal-voc-2012-183-labeledU2PL (DeepLab v3+ with ResNet-101)#14Validation mIoU: 69.2
semi-supervised-semantic-segmentation-on-pascal-voc-2012-732-labeledU2PL (DeepLab v3+ with ResNet-101)#14Validation mIoU: 76.2
semi-supervised-semantic-segmentation-on-cityscapes-12-5-labeledU2PL (DeepLab v3+ with ResNet-101 pretraind on ImageNet-1K, AEL)#15Validation mIoU: 76.48%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-92-labeledU2PL (DeepLab v3+ with ResNet-101)#15Validation mIoU: 68.0