Sample Prior Guided Robust Model Learning to Suppress Noisy Labels

Benchmark Model Rank Results
image-classification-on-cifar-10-with-noisyPGDF (ResNet-18)#3Accuracy (under 20% Sym. label noise): 96.7%
image-classification-on-clothing1mPGDF#8Accuracy: 75.19%
image-classification-on-mini-webvision-1-0PGDF (Inception-ResNet-v2)#2Top-1 Accuracy: 81.47Top-5 Accuracy: 94.03
learning-with-noisy-labels-on-cifar-100nPGDF#1Accuracy (mean): 74.08
learning-with-noisy-labels-on-cifar-10nPGDF#3Accuracy (mean): 96.11
learning-with-noisy-labels-on-cifar-10n-1PGDF#3Accuracy (mean): 96.01
learning-with-noisy-labels-on-cifar-10n-worstPGDF#3Accuracy (mean): 93.65