Debiased Learning from Naturally Imbalanced Pseudo-Labels

Benchmark Model Rank Results
few-shot-image-classification-on-imagenet-0DebiasPL (ResNet50)#1Accuracy: 68.3%
semi-supervised-image-classification-on-1DebiasPL (ResNet-50)#13Top 1 Accuracy: 71.3%
semi-supervised-image-classification-on-16DebiasPL (ResNet-50)#1ImageNet Top-1 Accuracy: 69.6%
semi-supervised-image-classification-on-cifar-6DebiasPL (w/ FixMatch)#2Percentage error: 4.6
semi-supervised-image-classification-on-cifar-7DebiasPL (w/ FixMatch)#7Percentage error: 5.4