Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

Overall framework of the SOS.

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.

More »

Fig 1 Expand

Fig 2.

The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 40%-asym.

More »

Fig 2 Expand

Fig 3.

The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 80%-sym.

More »

Fig 3 Expand

Fig 4.

The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 50%-sym.

More »

Fig 4 Expand

Fig 5.

The proportions of labeled and unlabeled samples divided by existing SOTA methods on CIFAR datasets with 80%-sym.

More »

Fig 5 Expand

Fig 6.

Visualizing samples from CIFAR-10.

More »

Fig 6 Expand

Fig 7.

Visualizing samples from CIFAR-100.

More »

Fig 7 Expand

Table 1.

Summary of datasets.

More »

Table 1 Expand

Table 2.

Settings of the hyperparameters used in this study.

More »

Table 2 Expand

Fig 8.

Visualizing samples from WebVision.

More »

Fig 8 Expand

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.

More »

Fig 9 Expand

Fig 10.

Curve of test accuracy on CIFAR-10.

The test results of SOS on CIFAR-10 with different noise rates.

More »

Fig 10 Expand

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.

More »

Fig 11 Expand

Fig 12.

Confusion matrix of SOS on the test set of CIFAR-10 when trained with 20%-sym.

scenario.

More »

Fig 12 Expand

Fig 13.

Confusion matrix of SOS on the test set of CIFAR-10 when trained with 50%-sym.

scenario.

More »

Fig 13 Expand

Fig 14.

Confusion matrix of SOS on the test set of CIFAR-10 when trained with 80%-sym.

scenario.

More »

Fig 14 Expand

Fig 15.

Confusion matrix of SOS on the test set of CIFAR-10 when trained with 90%-sym.

scenario.

More »

Fig 15 Expand

Fig 16.

Confusion matrix of SOS on the test set of CIFAR-10 when trained with 10%-asym.

scenario.

More »

Fig 16 Expand

Fig 17.

Confusion matrix of SOS on the test set of CIFAR-10 when trained with 30%-asym.

scenario.

More »

Fig 17 Expand

Fig 18.

Confusion matrix of SOS on the test set of CIFAR-10 when trained with 40%-asym.

scenario.

More »

Fig 18 Expand

Table 3.

Results on CIFAR-10 using PreAct ResNet-18.

More »

Table 3 Expand

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.

More »

Fig 19 Expand

Table 4.

Classification performance on CIFAR-10 with heavy symmetric noise.

More »

Table 4 Expand

Fig 20.

Curve of test accuracy on CIFAR-100.

More »

Fig 20 Expand

Fig 21.

The comparison between SOS and UNICON.

More »

Fig 21 Expand

Table 5.

Results on CIFAR-100.

More »

Table 5 Expand

Fig 22.

Comparison of test accuracy on CIFAR-10N.

More »

Fig 22 Expand

Fig 23.

The comparison on CIFAR-100N.

More »

Fig 23 Expand

Table 6.

Results on CIFAR-N.

More »

Table 6 Expand

Table 7.

Results on WebVision using pre-trained ResNet-50.

More »

Table 7 Expand

Table 8.

Results on Clothing1M using pre-trained ResNet-50.

More »

Table 8 Expand

Table 9.

Ablation results of SOS.

More »

Table 9 Expand

Table 10.

The training time cost (hours, i.e., “h”) on CIFAR-10 with 50% symmetric noise.

More »

Table 10 Expand