ClassMix: Segmentation-Based Data Augmentation for Semi-Supervised Learning

Benchmark Model Rank Results
semi-supervised-semantic-segmentation-on-pascal-voc-2012-1-labeledClassMix (DeepLab v2 MSCOCO pretrained)#5Validation mIoU: 54.18%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-2-labeledClassMix (DeepLab v2 MSCOCO pretrained)#6Validation mIoU: 66.15%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-5-labeledClassMix (DeepLab v2 MSCOCO pretrained)#8Validation mIoU: 67.77%
semi-supervised-semantic-segmentation-on-cityscapes-100-samples-labeledClassMix (DeepLab v2 MSCOCO pretrained)#11Validation mIoU: 54.07%
semi-supervised-semantic-segmentation-on-cityscapes-50-labeledClassMix (DeepLab v2 MSCOCO pretrained)#21Validation mIoU: 66.29%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-25-labeledClassMix (DeepLab v2 MSCOCO pretrained)#23Validation mIoU: 72.45
semi-supervised-semantic-segmentation-on-cityscapes-25-labeledClassMix (DeepLab v2 MSCOCO pretrained)#26Validation mIoU: 63.63%
semi-supervised-semantic-segmentation-on-cityscapes-12-5-labeledClassMix (DeepLab v2 MSCOCO pretrained)#28Validation mIoU: 61.35%
semi-supervised-semantic-segmentation-on-pascal-voc-2012-12-5-labeledClassMix#28Validation mIoU: 71.00%