List your company
ResearchFormerly WizWand

Image Classification on MIT67

81.1AccuracyRényiCL
34.913646.904358.89570.8857Feb 5, 2021May 8, 2021Aug 8, 2021Nov 8, 2021Feb 8, 2022May 11, 2022Aug 12, 2022
Updated 8mo ago

Evaluation results

MethodLinks
RényiCLEvaluation Protocol=linear evaluation, Multi-crop Augmentation=true2022.0881.1
CLSAEvaluation Protocol=linear evaluation, Multi-crop Augmentation=true2022.0877.7
CLSAEvaluation Protocol=linear evaluation, Multi-crop Augmentation=false2022.0877.1
InfoMinEvaluation Protocol=linear evaluation, Multi-crop Augmentation=false2022.0876.9
RényiCLEvaluation Protocol=linear evaluation, Multi-crop Augmentation=false2022.0875.8
AFDTeacher backbone=ResNet34, Student backbone=ResNet18, Pre-trained dataset=ImageNet, Evaluation protocol=fine-tuning2021.0275.07
L2TTeacher backbone=ResNet34, Student backbone=ResNet18, Pre-trained dataset=ImageNet, Evaluation protocol=fine-tuning2021.0274.85
ATTTeacher backbone=ResNet34, Student backbone=ResNet18, Pre-trained dataset=ImageNet, Evaluation protocol=fine-tuning2021.0272.54
FitNetTeacher backbone=ResNet34, Student backbone=ResNet18, Pre-trained dataset=ImageNet, Evaluation protocol=fine-tuning2021.0271.72
Fine-tuningTeacher backbone=ResNet34, Student backbone=ResNet18, Pre-trained dataset=ImageNet, Evaluation protocol=fine-tuning2021.0270.67
SimSiam + AugSelfFine-tuning strategy=L2SP, Pre-trained on=ImageNet100, Backbone=ResNet-502021.1169.31
SimSiam + AugSelfBackbone=ResNet-50, Pre-training Dataset=ImageNet100, Evaluation Protocol=Linear evaluation2021.1167.76
SimSiamFine-tuning strategy=L2SP, Pre-trained on=ImageNet100, Backbone=ResNet-502021.1167.69
AFDTeacher architecture=ResNet34, Student architecture=ResNet18, Source domain=ImageNet, Number of experiment repeats=3, Beta=1,0002021.0266.47
SimSiamBackbone=ResNet-50, Pre-training Dataset=ImageNet100, Evaluation Protocol=Linear evaluation2021.1165.75
L2TTeacher architecture=ResNet34, Student architecture=ResNet18, Source domain=ImageNet, Number of experiment repeats=32021.0264.85
MoCo v2 + AugSelfBackbone=ResNet-50, Pre-training Dataset=ImageNet100, Evaluation Protocol=Linear evaluation2021.1163.36
MoCo v2Backbone=ResNet-50, Pre-training Dataset=ImageNet100, Evaluation Protocol=Linear evaluation2021.1161.64
ATTTeacher architecture=ResNet34, Student architecture=ResNet18, Source domain=ImageNet, Number of experiment repeats=32021.0259.18
FitNetTeacher architecture=ResNet34, Student architecture=ResNet18, Source domain=ImageNet, Number of experiment repeats=32021.0254.88
Supervised + AugSelfBackbone=ResNet-50, Pre-training Dataset=ImageNet100, Evaluation Protocol=Linear evaluation2021.1154.63
SupervisedBackbone=ResNet-50, Pre-training Dataset=ImageNet100, Evaluation Protocol=Linear evaluation2021.1152.91
ScratchTeacher architecture=ResNet34, Student architecture=ResNet18, Source domain=ImageNet, Number of experiment repeats=32021.0248.91
BYOLAugSelf=true2021.1146.17
SimSiam + AugSelfBackbone=ResNet-18, Pre-training Dataset=STL10, Evaluation Protocol=Linear evaluation2021.1145.67
BYOLAugSelf=false2021.1142.78
MoCo v2 + AugSelfBackbone=ResNet-18, Pre-training Dataset=STL10, Evaluation Protocol=Linear evaluation2021.1141.67
SimCLRAugSelf=true2021.1140.62
SimSiamBackbone=ResNet-18, Pre-training Dataset=STL10, Evaluation Protocol=Linear evaluation2021.1139.15
SwAVAugSelf=true2021.1139.13
MoCo v2Backbone=ResNet-18, Pre-training Dataset=STL10, Evaluation Protocol=Linear evaluation2021.1139.01
SimCLRAugSelf=false2021.1136.82
SwAVAugSelf=false2021.1136.69