Semi-supervised semantic segmentation needs strong, varied perturbations

Benchmark Model Rank Results
semi-supervised-semantic-segmentation-on-1CutMix (DeepLab v2, ImageNet pre-trained)#25Validation mIoU: 63.87%
semi-supervised-semantic-segmentation-on-2CutMix (DeepLab v2, ImageNet pre-trained)#29Validation mIoU: 60.34%
semi-supervised-semantic-segmentation-on-23CutMix-Seg (Range View)#5mIoU (1% Labels): 36.7mIoU (10% Labels): 50.7mIoU (20% Labels): 52.9
semi-supervised-semantic-segmentation-on-24CutMix-Seg (Range View)#9mIoU (1% Labels): 37.4mIoU (10% Labels): 54.3mIoU (20% Labels): 56.6
semi-supervised-semantic-segmentation-on-25CutMix-Seg (Range View)#7mIoU (1% Labels): 43.8mIoU (10% Labels): 63.9mIoU (20% Labels): 64.8
semi-supervised-semantic-segmentation-on-3CutMix (DeepLab v2, ImageNet pre-trained)#12Validation mIoU: 51.2
semi-supervised-semantic-segmentation-on-4CutMix#25Validation mIoU: 72.45%
semi-supervised-semantic-segmentation-on-4CutMix#31Validation mIoU: 67.6%
semi-supervised-semantic-segmentation-on-41CutMix#5Validation mIoU: 26.2
semi-supervised-semantic-segmentation-on-42CutMix#5Validation mIoU: 29.8
semi-supervised-semantic-segmentation-on-5CutMix (DeepLab v3+ ImageNet pre-trained)#6Validation mIoU: 69.57%
semi-supervised-semantic-segmentation-on-5CutMix (DeepLab v2 ImageNet pre-trained)#11Validation mIoU: 66.48%
semi-supervised-semantic-segmentation-on-6CutMix (DeepLab v3+ ImageNet pre-trained)#4Validation mIoU: 67.05%
semi-supervised-semantic-segmentation-on-6CutMix (DeepLab v2 ImageNet pre-trained)#7Validation mIoU: 64.81%
semi-supervised-semantic-segmentation-on-7CutMix (DeepLab v3+ ImageNet pre-trained)#4Validation mIoU: 59.52%
semi-supervised-semantic-segmentation-on-7CutMix (DeepLab v2 ImageNet pre-trained)#6Validation mIoU: 53.79%