Papers with Code · CC BY-SA 4.0
Semi-Supervised Image Classification
Semi-supervised image classification leverages unlabelled data as well as labelled data to increase classification performance.
You may want to read some blog posts to get an overview before reading the papers and checking the leaderboards:
- An overview of proxy-label approaches for semi-supervised learning - Sebastian Ruder
- Semi-Supervised Learning in Computer Vision - Amit Chaudhary
<span style="color:grey; opacity: 0.6">( Image credit: Self-Supervised Semi-Supervised Learning )</span>
데이터셋
70개ImageNet - 10% labeled dataImageNet - 1% labeled dataCIFAR-10, 4000 Labelscifar-100, 10000 LabelsCIFAR-10, 250 LabelsCIFAR-10, 40 LabelsCIFAR-100, 400 LabelsSVHN, 1000 labelsCIFAR-100, 2500 LabelsSVHN, 250 LabelsSTL-10, 1000 LabelsCIFAR-10, 1000 LabelsSVHN, 500 LabelsCIFAR-10ImageNet-100 (TEMI Split)SVHN, 40 LabelsCIFAR-10, 100 Labels (OpenSet, 6/4)CIFAR-10, 2000 LabelsCIFAR-10, 400 Labels (OpenSet, 6/4)CIFAR-10, 50 Labels (OpenSet, 6/4)Mini-ImageNet, 4000 LabelsSTL-10, 40 Labelscifar10, 250 LabelsCIFAR-10, 20 LabelsCIFAR-100ImageNet - 0.2% labeled dataMini-ImageNet, 1000 LabelsMini-ImageNet, 10000 LabelsSTL-10cifar-10, 10 Labels