Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results

Benchmark Model Rank Results
semi-supervised-image-classification-on-2Mean Teacher (ResNeXt-152)#46Top 5 Accuracy: 90.89%
semi-supervised-image-classification-on-cifarMean Teacher#22Percentage error: 6.28
semi-supervised-image-classification-on-cifar-6MeanTeacher#16Percentage error: 47.32
semi-supervised-image-classification-on-svhnMean Teacher#13Accuracy: 96.05
semi-supervised-image-classification-on-svhn-1MeanTeacher#11Accuracy: 93.55
semi-supervised-semantic-segmentation-on-23MeanTeacher (Voxel)#2mIoU (1% Labels): 41.0mIoU (10% Labels): 50.1mIoU (20% Labels): 52.8
semi-supervised-semantic-segmentation-on-23MeanTeacher (Range View)#7mIoU (1% Labels): 34.2mIoU (10% Labels): 49.8mIoU (20% Labels): 51.6
semi-supervised-semantic-segmentation-on-24MeanTeacher (Voxel)#5mIoU (1% Labels): 45.4mIoU (10% Labels): 57.1mIoU (20% Labels): 59.2
semi-supervised-semantic-segmentation-on-24MeanTeacher (Range View)#8mIoU (1% Labels): 37.5mIoU (10% Labels): 53.1mIoU (20% Labels): 56.1
semi-supervised-semantic-segmentation-on-25MeanTeacher (Voxel)#4mIoU (1% Labels): 51.6mIoU (10% Labels): 66.0mIoU (20% Labels): 67.1
semi-supervised-semantic-segmentation-on-25MeanTeacher (Range View)#8mIoU (1% Labels): 42.1mIoU (10% Labels): 60.4mIoU (20% Labels): 65.4
source-free-object-detection-on-cityscapes-toMT#8AP50: 34.3