Semi-Supervised Semantic Segmentation with Pixel-Level Contrastive Learning from a Class-wise Memory Bank

Benchmark Model Rank Results
semi-supervised-semantic-segmentation-on-cityscapes-100-samples-labeledSemiSegContrast (DeepLab v3+ with ResNet-50 backbone, MSCOCO pretrained)#4Validation mIoU: 64.9%
semi-supervised-semantic-segmentation-on-cityscapes-100-samples-labeledSemiSegContrast (DeepLab v2 with ResNet-101 backbone, MSCOCO pretrained)#7Validation mIoU: 59.4%
semi-supervised-semantic-segmentation-on-cityscapes-12-5-labeledSemiSegContrast (DeepLab v2 with ResNet-101 backbone, MSCOCO pretrained)#26Validation mIoU: 64.4%
semi-supervised-semantic-segmentation-on-cityscapes-25-labeledSemiSegContrast (DeepLab v2 with ResNet-101 backbone, MSCOCO pretrained)#24Validation mIoU: 65.9%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-12-5-labeledSemiSegContrast#26Validation mIoU: 71.6%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-2-labeledSemiSegContrast (DeepLab v2 with ResNet-101 backbone, MSCOCO pretrained)#2Validation mIoU: 67.9%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-5-labeledSemiSegContrast (DeepLab v2 with ResNet-101 backbone, MSCOCO pretrained)#4Validation mIoU: 70.0%