List your company
ResearchFormerly WizWand

Image Classification Top-1 Accuracy on CIFAR100 (test)

93.4Top-1 AccuracyWiSE-FT
76.666481.010785.35589.6993Jun 18, 2020Apr 4, 2021Jan 19, 2022Nov 6, 2022Aug 23, 2023Jun 8, 2024Mar 26, 2025
Updated 7mo ago

Evaluation results

MethodLinks
WiSE-FTalpha=optimal2021.0993.4
WiSE-FTalpha=0.52021.0993.3
TResNet-LPre-training Dataset=ImageNet-21K-P2021.0492.5
Standard fine-tuning2021.0992.2
TResNet-MPre-training Dataset=ImageNet-21K-P2021.0491.7
OFA595Pre-training Dataset=ImageNet-21K-P2021.0490.3
ResNet50Pre-training Dataset=ImageNet-21K-P2021.0490.3
TResNet-LPre-training Dataset=ImageNet-1K2021.0490.2
TResNet-MPre-training Dataset=ImageNet-1K2021.0489.5
MobileNetV3Pre-training Dataset=ImageNet-21K-P2021.0488.5
OFA595Pre-training Dataset=ImageNet-1K2021.0488.3
ResNet50Pre-training Dataset=ImageNet-1K2021.0486.8
MobileNetV3Pre-training Dataset=ImageNet-1K2021.0486.7
Supervised-INEvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0186.4
BYOLEvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0186.1
SimCLREvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0185.9
ReLICv2Evaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=ImageNet2022.0185.3
CDARTSParam (M)=3.9 ± 0.08, Latency (ms)=41.2 ± 0.5, Search Cost (GPU days)=0.3, Search Method=gradient, Search space=DARTS search space2020.0684.31
PDARTSParam (M)=3.4, Latency (ms)=40.9, Search Cost (GPU days)=0.3, Search Method=gradient, Search space=DARTS search space2020.0683.45
DenseNet-BCParam (M)=25.6, Search Method=manual2020.0682.82
Wide ResNetParam (M)=36.5, Search Method=manual2020.0682.7
ResNeXt-29, 16x64dParam (M)=68.1, Search Method=manual2020.0682.69
DARTSV2Latency (ms)=3.3, Search Cost (GPU days)=4.0, Search Method=gradient2020.0682.46
DARTSV1Param (M)=3.3, Search Cost (GPU days)=1.5, Search Method=gradient2020.0682.24
GDASParam (M)=3.4, Latency (ms)=30.6, Search Cost (GPU days)=4, Search Method=gradient2020.0681.62
NAONetParam (M)=10.6, Search Cost (GPU days)=200, Search Method=NAO2020.0681.33
SAMNetwork=ResNet101, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0380.82
CCSSL (FixMatch)Number of Labeled Samples=100002022.0380.68
FLSDModel=WideResNet2025.0380.44
RMSGDNetwork=ResNet101, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0380.36
DFLModel=WideResNet2025.0380.35
BSCE-GRAModel=WideResNet2025.0380.28
Random InitEvaluation protocol=Fine-tuned, Backbone=ResNet-50, Pre-trained=None2022.0180.2
RMSGDNetwork=ResNet50, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0380.06
SGDPNetwork=ResNet101, Optimizer=SGDP, Cutout=true, Batch size=128, Epochs=2502022.0380.03
BSCEModel=WideResNet2025.0379.96
SAMNetwork=ResNet34, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0379.85
RMSGDNetwork=ResNet34, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0379.7
SGDPNetwork=ResNet34, Optimizer=SGDP, Cutout=true, Batch size=128, Epochs=2502022.0379.67
RMSGDNetwork=ResNet-50, Cutout=No2022.0379.59
SGDPNetwork=ResNet50, Optimizer=SGDP, Cutout=true, Batch size=128, Epochs=2502022.0379.52
CEModel=WideResNet2025.0379.51
RMSGDNetwork=ResNet-101, Cutout=No2022.0379.36
SGDNetwork=ResNet101, Optimizer=SGD, Cutout=true, Batch size=128, Epochs=2502022.0379.35
RMSGDNetwork=ResNet-34, Cutout=No2022.0379.32
SAMNetwork=ResNet50, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0379.26
CoMatchNumber of Labeled Samples=100002022.0379.14
MMCEModel=WideResNet2025.0379.14
SSWPLNumber of Labeled Samples=100002022.0379.12
ResNet-34 vanillaBackbone=ResNet-34, Model DoF=21M2021.0779.1
R34-vanillaDoF=21M2021.0779.1
NNCLREvaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0179
ResNet-34 RPGBackbone=ResNet-34, Model DoF=11M2021.0778.9
R34-RPGDoF=11M2021.0778.9
SAMNetwork=SENet18, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0378.85
SGDPNetwork=ResNet18, Optimizer=SGDP, Cutout=true, Batch size=128, Epochs=2502022.0378.82
SAMNetwork=ResNet18, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0378.77
SGDPNetwork=ResNet-34, Cutout=No2022.0378.74
BLModel=WideResNet2025.0378.72
SGDNetwork=ResNet34, Optimizer=SGD, Cutout=true, Batch size=128, Epochs=2502022.0378.63
RMSGDNetwork=ResNet-18, Cutout=No2022.0378.63
SGDPNetwork=ResNet-101, Cutout=No2022.0378.6
RMSGDNetwork=ResNet18, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0378.53
R34-RPG.blkDoF=11M2021.0778.5
SEALGeneration=Gen=Last (Student), Backbone=ResNet-182023.0478.5
SGDNetwork=ResNet-101, Cutout=No2022.0378.48
SGDPNetwork=ResNet-50, Cutout=No2022.0378.44
R34-LegoDoF=11M2021.0778.4
BYOLEvaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0178.4
SAMNetwork=ResNet-101, Cutout=No2022.0378.38
SGDNetwork=ResNet50, Optimizer=SGD, Cutout=true, Batch size=128, Epochs=2502022.0378.36
Supervised-INEvaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0178.3
ReLICv2Evaluation protocol=Linear evaluation, Backbone=ResNet-50, Pre-trained=ImageNet2022.0178.2
SGDNetwork=ResNet18, Optimizer=SGD, Cutout=true, Batch size=128, Epochs=2502022.0378.16
SGDPNetwork=ResNet-18, Cutout=No2022.0378.13
SGDNetwork=ResNet-50, Cutout=No2022.0378.12
RMSGDNetwork=ResNeXt, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0378.06
Teacher (Scratch)Architecture=ResNet34/ResNet182022.0978.05
SAMNetwork=ResNet-34, Cutout=No2022.0377.98
SGDNetwork=ResNet-34, Cutout=No2022.0377.88
SGDNetwork=SENet18, Optimizer=SGD, Cutout=true, Batch size=128, Epochs=2502022.0377.8
SGDNetwork=ResNet-18, Cutout=No2022.0377.8
RMSGDNetwork=SENet18, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0377.77
FLSDModel=ResNet1102025.0377.77
AdamPNetwork=ResNet-101, Cutout=No2022.0377.71
SGDPNetwork=SENet18, Optimizer=SGDP, Cutout=true, Batch size=128, Epochs=2502022.0377.7
FLSDModel=ResNet502025.0377.69
ResNet-18 RPGBackbone=ResNet-18, Model DoF=5.5M2021.0777.6
ResNet-18 vanillaBackbone=ResNet-18, Model DoF=11M2021.0777.6
R18-vanillaDoF=11M2021.0777.5
MMCEModel=ResNet502025.0377.49
AdamPNetwork=ResNet-50, Cutout=No2022.0377.47
CEModel=ResNet1102025.0377.44
MMCEModel=ResNet1102025.0377.42
SAMNetwork=ResNeXt, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0377.41
FixMatchNumber of Labeled Samples=100002022.0377.4
SAMNetwork=ResNet-50, Cutout=No2022.0377.39
float32Model=ResNet18, Bit-width=float32, Model Size=43 MB2021.0377.38
Pham et al.Generation=Gen=Last (Student), Backbone=ResNet-182023.0477.32
MADArchitecture=ResNet34/ResNet182022.0977.31