Fig 1.
Each network consists of an extractor g(⋅), a classifier f(⋅), and a projection header h(⋅), similar to simCLR [52]. During training, the dataset is divided into a labeled subset and an unlabeled subset through uniform selection, and then the useful information in the label-free samples is extracted by combining the oversampling strategy with a robust training process, where the robust training is performed using the SSL training technique MixMatch and unsupervised CL. In each network, the two subsets, derived separately from the two networks, are utilized for training. The training process is cyclic, involving repeated iterations of these steps.
Fig 2.
The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 40%-asym.
Fig 3.
The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 80%-sym.
Fig 4.
The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 50%-sym.
Fig 5.
The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 80%-sym.
Fig 6.
Visualizing samples from CIFAR-10.
Fig 7.
Visualizing samples from CIFAR-100.
Table 1.
Summary of datasets.
Table 2.
Settings of the hyperparameters used in this study.
Fig 8.
Visualizing samples from WebVision.
Fig 9.
Visualizing samples from Clothing1M.
We randomly select 10 images from each of the first 10 categories for display from CIFAR datasets and 14 images from each of the first 14 categories from WebVision and Clothing1M.
Fig 10.
Curve of test accuracy on CIFAR-10.
The test results of SOS on CIFAR-10 with different noise rates.
Fig 11.
The comparison of classification performance between SOS and UNICON.
The steep drop in two figures means the end of the warm-up stage.
Fig 12.
Confusion matrix of SOS on the test set of CIFAR-10 when trained with 20%-sym.
scenario.
Fig 13.
Confusion matrix of SOS on the test set of CIFAR-10 when trained with 50%-sym.
scenario.
Fig 14.
Confusion matrix of SOS on the test set of CIFAR-10 when trained with 80%-sym.
scenario.
Fig 15.
Confusion matrix of SOS on the test set of CIFAR-10 when trained with 90%-sym.
scenario.
Fig 16.
Confusion matrix of SOS on the test set of CIFAR-10 when trained with 10%-asym.
scenario.
Fig 17.
Confusion matrix of SOS on the test set of CIFAR-10 when trained with 30%-asym.
scenario.
Fig 18.
Confusion matrix of SOS on the test set of CIFAR-10 when trained with 40%-asym.
scenario.
Table 3.
Results on CIFAR-10 using PreAct ResNet-18.
Fig 19.
Test curve of SOS and UNICON on CIFAR-10 with severe label noise.
“SOS (30)” and “SOS (50)” represent the results of our method when λu = 30 and λu = 50, respectively.
Table 4.
Classification performance on CIFAR-10 with heavy symmetric noise.
Fig 20.
Curve of test accuracy on CIFAR-100.
Fig 21.
The comparison between SOS and UNICON.
Table 5.
Results on CIFAR-100.
Fig 22.
Comparison of test accuracy on CIFAR-10N.
Fig 23.
The comparison on CIFAR-100N.
Table 6.
Results on CIFAR-N.
Table 7.
Results on WebVision using pre-trained ResNet-50.
Table 8.
Results on Clothing1M using pre-trained ResNet-50.
Table 9.
Ablation results of SOS.
Table 10.
The training time cost (hours, i.e., “h”) on CIFAR-10 with 50% symmetric noise.