HyperAI超神经

Semi Supervised Image Classification

半监督图像分类是一种结合有标签数据和无标签数据以提升分类性能的技术。该方法通过利用大量未标注图像,增强模型的泛化能力和准确性,有效缓解标注数据不足的问题,在计算机视觉领域具有重要应用价值。

Caltech-101
Caltech-101, 202 Labels
Caltech-256
Caltech-256, 1024 Labels
cifar-10, 10 Labels
BOSS
CIFAR-10, 100 Labels
SimCLR-kmediods-PAWS
CIFAR-10, 100 Labels (OpenSet, 6/4)
CIFAR-10, 1000 Labels
MixMatch
CIFAR-10, 20 Labels
CIFAR-10, 2000 Labels
MixMatch
CIFAR-10, 250 Labels
CIFAR-10 (250 Labels, ImageNet-100 Unlabeled)
CIFAR-10, 30 Labels
CIFAR-10, 40 Labels
FreeMatch
CIFAR-10, 400 Labels (OpenSet, 6/4)
UnMixMatch
CIFAR-10, 4000 Labels
CIFAR-10 (4000 Labels, ImageNet-100 Unlabeled)
CIFAR-10, 50 Labels (OpenSet, 6/4)
CIFAR-10, 500 Labels
MixMatch
CIFAR-10, 80 Labels
SimCLR (CoMatch)
CIFAR-100, 1000 Labels
EnAET
cifar-100, 10000 Labels
CCSSL(FixMatch)
CIFAR-100 (10000 Labels, ImageNet-100 Unlabeled)
CIFAR-100, 200 Labels
CIFAR-100 (250 Labels, ImageNet-100 Unlabeled)
CCSSL
CIFAR-100, 2500 Labels
FlexMatch
CIFAR-100, 400 Labels
SemiReward
CIFAR-100 (400 Labels, ImageNet-100 Unlabeled)
CIFAR-100, 4000 Labels
UPS (CNN-13)
CIFAR-100, 5000 Labels
CIFAR-100, 5000Labels
EnAET
cifar10, 250 Labels
ReMixMatch
DeepWeeds, 99 Labels
EuroSAT, 100 Labels
EuroSAT, 20 Labels
SimCLR-kmediods-PAWS
ImageNet - 0.2% labeled data
DebiasPL (ResNet-50)
ImageNet - 1% labeled data
REACT (ViT-Large)
ImageNet - 10% labeled data
Meta Co-Training
Imagenette, 100 Labels
Imagenette, 20 Labels
Mini-ImageNet, 1000 Labels
MutexMatch
Mini-ImageNet, 10000 Labels
FeatMatch
Mini-ImageNet, 4000 Labels
SimPLE
Salinas
Res-CP
STL-10
EnAET
STL-10, 1000 Labels
Semi-MMDC
STL-10 (1000 Labels, ImageNet-100 Unlabeled)
STL-10, 40 Labels
RelationMatch
STL-10, 5000 Labels
MixMatch
SVHN, 1000 labels
Meta Pseudo Labels (WRN-28-2)
SVHN (1000 Labels, ImageNet-100 Unlabeled)
SVHN, 2000 Labels
MixMatch
SVHN, 250 Labels
EnAET
SVHN (250 Labels, ImageNet-100 Unlabeled)
SVHN, 40 Labels
ShrinkMatch
SVHN (40 Labels, ImageNet-100 Unlabeled)
SVHN, 4000 Labels
MixMatch
SVHN, 500 Labels
Triple-GAN-V2 (CNN-13)