List your company
ResearchFormerly WizWand

Semi-supervised Image Classification on ImageNet (10% labeled)

79Top-1 AccuracyPAWS
47.17655.43863.771.962Nov 23, 2020Jul 16, 2021Mar 8, 2022Oct 29, 2022Jun 21, 2023Feb 11, 2024Oct 3, 2024
Updated 8mo ago

Evaluation results

MethodLinks
PAWSArchitecture=ResNet-50 (4x), Epochs=2002021.0479—
PAWSArchitecture=ResNet-50 (4x), Epochs=1002021.0478.5—
PAWSArchitecture=ResNet-50 (2x), Epochs=2002021.0477.8—
PAWSArchitecture=ResNet-50 (2x), Epochs=1002021.0477—
BYOLArchitecture=ResNet-50 (4x), Epochs=10002021.0475.7—
SimCLRArchitecture=ResNet-50 (4x), Epochs=10002021.0474.4—
SimCLRv2Architecture=ResNet-50 (2x), Epochs=8002021.0473.9—
BYOLArchitecture=ResNet-50 (2x), Epochs=10002021.0473.5—
FNCBackbone=ResNet-50, Learning Type=Representation Learning / Contrastive2020.1171.190.2
SwAVBackbone=ResNet-502020.1170.289.9
SwAVEpochs=800, Backbone=ResNet-502024.1070.289.9
BTEpochs=1000, Backbone=ResNet-502024.1069.789.3
BYOLBackbone=ResNet-502020.1168.889
BYOLEpochs=1000, Backbone=ResNet-502024.1068.889
SimCLR v2Backbone=ResNet-50, Learning Type=Representation Learning / Contrastive2020.1168.489.2
SynCoEpochs=800, Backbone=ResNet-502024.1066.688
MoCo-v2 (repr.)Epochs=800, Backbone=ResNet-502024.1066.187.6
MOCHI (repr.)Epochs=800, Backbone=ResNet-502024.1065.787.2
SimCLR v1Backbone=ResNet-50, Learning Type=Representation Learning / Contrastive2020.1165.687.8
DATA (7G~8G)Params=45.2M, Fine-tuning epochs=202022.0365.686.9
SimCLREpochs=1000, Backbone=ResNet-502024.1065.687.8
DATA (6G~7G)Params=45.4M, Fine-tuning epochs=202022.0365.487
DATA (5G~6G)Params=43.4M, Fine-tuning epochs=202022.0365.186.7
DATA (4G~5G)Params=37.0M, Fine-tuning epochs=202022.0364.386.4
DATA (3G~4G)Params=33.5M, Fine-tuning epochs=202022.0363.485.7
DATA (2G~3G)Params=19.5M, Fine-tuning epochs=202022.036385.8
DATA (1G~2G)Params=14.7M, Fine-tuning epochs=202022.0361.984.9
ResNet-50Params=25.5M, Pre-training=MoCo v2, Fine-tuning epochs=202022.0361.885.1
PIRLEpochs=800, Backbone=ResNet-502024.1060.483.8
PIRLBackbone=ResNet-50, Learning Type=Representation Learning / Contrastive2020.1157.283.8
SupervisedBackbone=ResNet-502024.1056.480.4
SupervisedBackbone=ResNet-502020.1148.480.4
FixMatchBackbone=ResNet-50, Learning Type=Semi-supervised2020.11—89.1
PCLBackbone=ResNet-50, Learning Type=Representation Learning / Contrastive2020.11—86.2
UDABackbone=ResNet-50, Learning Type=Semi-supervised2020.11—88.5