List your company
ResearchFormerly WizWand

Image Classification on CIFAR10 (test) (Accuracy)

99.5AccuracyWiSE-FT
96.109696.989897.8798.7502Jun 7, 2021Mar 19, 2022Dec 29, 2022Oct 10, 2023Jul 21, 2024May 2, 2025Feb 12, 2026
Updated 8mo ago

Evaluation results

MethodLinks
WiSE-FTalpha=optimal2021.0999.5——
WiSE-FTalpha=0.52021.0999.3——
DeiT-BParams (M)=86.6, Fine-tuning=true2021.0699.1——
ViT-baseImplementation=Original2022.1298.95——
ViTAE-SParams (M)=23.6, Fine-tuning=true2021.0698.8——
ViT-baseImplementation=TencentPretrain2022.1298.73——
Standard fine-tuning2021.0998.6——
ARFCBackbone=SigLip 2 (ViT-B/16), Compression Ratio=50%, Evaluation Protocol=Zero-shot2026.0298.5——
BaselineBackbone=SigLip 2 (ViT-B/16), Compression Ratio=0%, Evaluation Protocol=Zero-shot2026.0298.4——
T2T-ViT-14Params (M)=21.5, Fine-tuning=true2021.0698.3——
EfficientNet-B5Params (M)=30, Fine-tuning=true2021.0698.1——
ViT-B/16Params (M)=86.5, Fine-tuning=true2021.0698.1——
Clust-PSI-PFLPartition protocol=Similarity, S=0, K=1002025.1298——
VaSSOBackbone=PyramidNet-1102023.0997.93——
ViT-L/16Params (M)=304.3, Fine-tuning=true2021.0697.9——
AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), Compression Ratio=50%, Evaluation Protocol=Zero-shot2026.0297.9——
Invariance-Constrained LearningBackbone=Wide-ResNet-28-10, Augmentation Search Space=Wide2022.0997.85——
SAMBackbone=PyramidNet-1102023.0997.85——
FisherSAMBackbone=PyramidNet-1102023.0997.84——
BYOLEvaluation Protocol=Fine-tuned2021.0697.8——
SSL-HSIC (w/ target)Evaluation Protocol=Fine-tuned2021.0697.8——
Invariance-Constrained LearningBackbone=Wide-ResNet-28-10, Augmentation Search Space=Standard2022.0997.74——
SimCLREvaluation Protocol=Fine-tuned2021.0697.7——
ViTAE-TParams (M)=4.8, Epochs=1000, Training mode=from scratch2021.0697.7——
DeepAABackbone=Wide-ResNet-28-10, Augmentation Search Space=Standard2022.0997.57——
DeepAABackbone=WRN-28-10, Total Augmentations=6, Hard-coded Augmentations=02023.0697.56——
ASAMBackbone=PyramidNet-1102023.0997.56——
VaSSOBackbone=WRN-28-102023.0997.54——
TABackbone=Wide-ResNet-28-10, Augmentation Search Space=Wide2022.0997.51——
Supervised-INEvaluation Protocol=Fine-tuned2021.0697.5——
DeiT-BParams (M)=86.0, Epochs=7000, Training mode=from scratch2021.0697.5——
TeachABackbone=WRN-28-10, Total Augmentations=4, Hard-coded Augmentations=22023.0697.5——
Q-FormerBackbone=SigLip 2 (ViT-B/16), Compression Ratio=50%, Evaluation Protocol=Zero-shot2026.0297.5——
TA (Wide)Backbone=WRN-28-10, Total Augmentations=3, Hard-coded Augmentations=22023.0697.46——
SLACKBackbone=WRN-28-10, Total Augmentations=3, Hard-coded Augmentations=02023.0697.46——
TABackbone=Wide-ResNet-28-10, Augmentation Search Space=Standard2022.0997.46——
FisherSAMBackbone=WRN-28-102023.0997.46——
SparseSwin#Params(10 class)=17.58 M, Input Resolution=224x224, Model Type=Transformer2023.0997.43——
SSL-HSIC (w/o target)Evaluation Protocol=Fine-tuned2021.0697.4——
AABackbone=WRN-28-10, Total Augmentations=4, Hard-coded Augmentations=22023.0697.4——
SpikingResformerType=Transfer Learning, T=42024.0397.4——
SGDBackbone=PyramidNet-1102023.0997.39——
UniformAugmentBackbone=WRN-28-10, Total Augmentations=4, Hard-coded Augmentations=22023.0697.33——
SAMBackbone=WRN-28-102023.0997.32——
ViTAE-TParams (M)=4.8, Fine-tuning=true2021.0697.3——
FastAABackbone=WRN-28-10, Total Augmentations=4, Hard-coded Augmentations=22023.0697.3——
DADABackbone=WRN-28-10, Total Augmentations=4, Hard-coded Augmentations=22023.0697.3——
RABackbone=WRN-28-10, Total Augmentations=4, Hard-coded Augmentations=22023.0697.3——
DenseNet-BC-190+Mixup#Params(10 class)=25.6 M, Input Resolution=224x224, Model Type=Convolution2023.0997.3——
DeepAA (reproduced)Backbone=WRN-28-10, Total Augmentations=6, Hard-coded Augmentations=02023.0697.27——
DeepAABackbone=Wide-ResNet-28-10, Augmentation Search Space=Wide2022.0997.27——
Uniform policyBackbone=WRN-28-10, Total Augmentations=3, Hard-coded Augmentations=02023.0697.26——
Softmax + SAM2022.0997.2——
CRATEModel variant=L, # parameters=77.64M, Pre-training=ImageNet, Evaluation Protocol=Fine-tuning2023.0697.2——
ViTModel variant=S, # parameters=22.05M, Pre-training=ImageNet, Evaluation Protocol=Fine-tuning2023.0697.2——
NesT-B#Params(10 class)=97.2 M, Input Resolution=32x32, Model Type=Transformer2023.0997.2——
CRATE-L#Params(10 class)=77.64 M, Input Resolution=224x224, Model Type=Transformer2023.0997.2——
ASAMBackbone=WRN-28-102023.0997.15——
SAMNetwork=ResNet34, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0397.1——
SGDBackbone=WRN-28-102023.0997.08——
Invariance-Constrained LearningBackbone=Wide-ResNet-40-2, Augmentation Search Space=Wide2022.0997.05——
SpikformerType=Transfer Learning, T=42024.0397.03——
PTQ-P4Q (8bit)-FTBackbone=SigLip 2 (ViT-B/16), Compression Ratio=75%, Evaluation Protocol=Zero-shot2026.0296.9——
ResNet XnIDR#Params(10 class)=23.86 M, Input Resolution=224x224, Model Type=Convolution2023.0996.87——
CRATEModel variant=B, # parameters=22.80M, Pre-training=ImageNet, Evaluation Protocol=Fine-tuning2023.0696.8——
CRATE-B#Params(10 class)=22.80 M, Input Resolution=224x224, Model Type=Transformer2023.0996.8——
SAMNetwork=ResNet101, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0396.79——
VaSSOBackbone=ResNet-182023.0996.77——
Invariance-Constrained LearningBackbone=Wide-ResNet-40-2, Augmentation Search Space=Standard2022.0996.76——
FisherSAMBackbone=ResNet-182023.0996.72——
SGDPNetwork=ResNet34, Augmentation=Cutout2022.0396.7——
SGDPNetwork=ResNet34, Optimizer=SGDP, Cutout=true, Batch size=128, Epochs=2502022.0396.7——
SAMNetwork=ResNet34, Augmentation=Cutout2022.0396.64——
ViTModel variant=T, # parameters=5.72M, Pre-training=ImageNet, Evaluation Protocol=Fine-tuning2023.0696.6——
SAMBackbone=ResNet-182023.0996.58——
TABackbone=Wide-ResNet-40-2, Augmentation Search Space=Standard2022.0996.55——
SGDNetwork=ResNet34, Augmentation=Cutout2022.0396.53——
SGDNetwork=ResNet34, Optimizer=SGD, Cutout=true, Batch size=128, Epochs=2502022.0396.53——
SAMNetwork=ResNet50, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0396.49——
SAMNetwork=ResNet18, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0396.44——
SAMNetwork=SENet18, Optimizer=SAM, Cutout=true, Batch size=128, Epochs=2502022.0396.44——
DeepAABackbone=Wide-ResNet-40-2, Augmentation Search Space=Standard2022.0996.43——
RMSGDNetwork=ResNet34, Augmentation=Cutout2022.0396.42——
RMSGDNetwork=ResNet34, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0396.42——
FastAABackbone=WRN-40-2, Total Augmentations=4, Hard-coded Augmentations=22023.0696.4——
DADABackbone=WRN-40-2, Total Augmentations=4, Hard-coded Augmentations=22023.0696.4——
PTQ-MinMax (8bit)-FTBackbone=SigLip 2 (ViT-B/16), Compression Ratio=75%, Evaluation Protocol=Zero-shot2026.0296.4——
RMSGDNetwork=ResNet101, Augmentation=Cutout2022.0396.33——
RMSGDNetwork=ResNet101, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0396.33——
ASAMBackbone=ResNet-182023.0996.33——
TA (Wide)Backbone=WRN-40-2, Total Augmentations=3, Hard-coded Augmentations=22023.0696.32——
AABackbone=WRN-40-2, Total Augmentations=4, Hard-coded Augmentations=22023.0696.3——
SLACKBackbone=WRN-40-2, Total Augmentations=3, Hard-coded Augmentations=02023.0696.29——
RMSGDNetwork=ResNet50, Augmentation=Cutout2022.0396.28——
RMSGDNetwork=ResNet50, Optimizer=RMSGD, Cutout=true, Batch size=128, Epochs=2502022.0396.28——
DeepAABackbone=Wide-ResNet-40-2, Augmentation Search Space=Wide2022.0996.27——
UniformAugmentBackbone=WRN-40-2, Total Augmentations=4, Hard-coded Augmentations=22023.0696.25——
DeepAA (reproduced)Backbone=WRN-40-2, Total Augmentations=6, Hard-coded Augmentations=02023.0696.25——
SGDBackbone=ResNet-182023.0996.25——
TABackbone=Wide-ResNet-40-2, Augmentation Search Space=Wide2022.0996.24——