List your company
ResearchFormerly WizWand

Image Classification on ImageNet-W

-1.11IN-W GapMAE+LLE
-27.7236-20.8143-13.905-6.9957Dec 9, 2022
Updated 8mo ago

Evaluation results

MethodLinks
MAE+LLEtrain data=IN-1k, evaluation protocol=Last Layer Ensemble, backbone=ViT-H2022.12-1.11——————
MAE+LLEBackbone=ViT-L/16, Evaluation Protocol=Fine-tuned2022.12-1.74——————
MAE+LLEBackbone=ViT-B/16, Evaluation Protocol=Fine-tuned2022.12-2.48——————
SWAGtrain data=IG-3.6B, evaluation protocol=fine-tuning, backbone=ViT-H2022.12-3.09——————
SWAGArchitecture=ViT-H/14, (Pre)training data=IG-3.6B, Evaluation protocol=Fine-tuning2022.12-3.10.351820.25———
SWAGBackbone=ViT-H/14, Pre-training Data=IG-3.6B, Evaluation Protocol=Fine-tuning2022.12-3.1—18—90.3520.25
SWAGBackbone=ViT-L/16, Evaluation Protocol=Fine-tuned, Pre-training Dataset=IG-3.6B2022.12-3.16——————
SWAGBackbone=ViT-L/16, Pre-training Data=IG-3.6B, Evaluation Protocol=Fine-tuning2022.12-3.2—20—90.2419.14
MAEtrain data=IN-1k, evaluation protocol=fine-tuning, backbone=ViT-H2022.12-3.48——————
MAEArchitecture=ViT-H/14, (Pre)training data=IN-1k, Evaluation protocol=Fine-tuning2022.12-3.50.433029.59———
MAEBackbone=ViT-H/14, Pre-training Data=IN-1k, Evaluation Protocol=Fine-tuning2022.12-3.5—30—80.4329.59
CLIPArchitecture=ViT-H/14, (Pre)training data=LAION-2B, Evaluation protocol=zero-shot2022.12-3.60.031612.01———
CLIPBackbone=ViT-H/14, Pre-training Data=LAION-2B, Evaluation Protocol=Zero-shot2022.12-3.6—16—60.0312.01
CLIPtrain data=LAION-2B, evaluation protocol=zero-shot, backbone=ViT-H2022.12-3.61——————
CLIPBackbone=ViT-G/14, Pre-training Data=LAION-2B, Evaluation Protocol=Zero-shot2022.12-3.8—12—60.025.61
MAEBackbone=ViT-L/16, Evaluation Protocol=Fine-tuned2022.12-4.36——————
CLIPArchitecture=ViT-L/14, (Pre)training data=WIT, Evaluation protocol=zero-shot2022.12-4.40.01121.75———
MAEBackbone=ViT-L/16, Pre-training Data=IN-1k, Evaluation Protocol=Fine-tuning2022.12-4.4—22—90.522.7
CLIPBackbone=ViT-L/14, Pre-training Data=WIT, Evaluation Protocol=Zero-shot2022.12-4.4—12—60.011.75
CLIPBackbone=ViT-L/16, Evaluation Protocol=Zero-shot, Pre-training Dataset=WIT2022.12-4.47——————
SWAGBackbone=RG-32gf, Pre-training Data=IG-3.6B, Evaluation Protocol=Fine-tuning2022.12-4.5—30—80.4926.03
MAEBackbone=ViT-B/16, Evaluation Protocol=Fine-tuned2022.12-4.6——————
MAEBackbone=ViT-B/16, Pre-training Data=IN-1k, Evaluation Protocol=Fine-tuning2022.12-4.6—24—90.6722.46
SWAGtrain data=IG-3.6B, evaluation protocol=linear probing, backbone=ViT-H2022.12-4.89——————
CLIPArchitecture=ViT-L/14, (Pre)training data=LAION-400M, Evaluation protocol=zero-shot2022.12-4.90.031213.76———
SWAGArchitecture=ViT-H/14, (Pre)training data=IG-3.6B, Evaluation protocol=Linear Probing2022.12-4.90.19812.8———
CLIPBackbone=ViT-L/14, Pre-training Data=LAION-400M, Evaluation Protocol=Zero-shot2022.12-4.9—12—50.0313.76
SWAGBackbone=ViT-H/14, Pre-training Data=IG-3.6B, Evaluation Protocol=Linear Probing2022.12-4.9—8—90.1912.8
CLIPBackbone=ViT-L/16, Evaluation Protocol=Zero-shot, Pre-training Dataset=LAION2022.12-4.94——————
SWAGBackbone=ViT-B/16, Pre-training Data=IG-3.6B, Evaluation Protocol=Fine-tuning2022.12-5.4—24—90.4525.95
SWAGBackbone=ViT-B/16, Evaluation Protocol=Fine-tuned, Pre-training Dataset=IG-3.6B2022.12-5.43——————
SWAGBackbone=ViT-L/16, Pre-training Data=IG-3.6B, Evaluation Protocol=Linear Probing2022.12-5.7—6—80.239.72
SWAGBackbone=ViT-L/16, Evaluation Protocol=Linear Probing, Pre-training Dataset=IG-3.6B2022.12-5.73——————
ERMBackbone=ViT-L/162022.12-6.14——————
SupervisedArchitecture=ViT-L/16, (Pre)training data=IN-1k2022.12-6.20.823432.57———
SupervisedBackbone=ViT-L/16, Pre-training Data=IN-1k2022.12-6.2—34—80.8232.57
Greedy SoupBackbone=ViT-B/32, Evaluation Protocol=Fine-tuned, Pre-training Dataset=WIT2022.12-6.47——————
SEERArchitecture=RG-32gf, (Pre)training data=IG-1B, Evaluation protocol=Fine-tuning2022.12-6.50.561824.26———
Greedy SoupArchitecture=ViT-B/32, (Pre)training data=WIT, Evaluation protocol=Fine-tuning2022.12-6.50.351623.87———
SEERBackbone=RG-32gf, Evaluation Protocol=Fine-tuned, Pre-training Dataset=IG-1B2022.12-6.5——————
SEERBackbone=RG-32gf, Pre-training Data=IG-1B, Evaluation Protocol=Fine-tuning2022.12-6.5—18—90.5624.26
SWAGBackbone=RG-32gf, Pre-training Data=IG-3.6B, Evaluation Protocol=Linear Probing2022.12-6.5—22—80.3620.56
Greedy SoupBackbone=ViT-B/32, Pre-training Data=WIT, Evaluation Protocol=Fine-tuning2022.12-6.5—16—90.3523.87
ERMBackbone=ViT-B/162022.12-6.69——————
SupervisedBackbone=ViT-B/16, Pre-training Data=IN-1k2022.12-6.7—26—80.7331.28
RobustViTBackbone=ViT-B/16, Pre-training Data=IN-1k2022.12-7.3—34—80.4437.06
SWAGBackbone=ViT-B/16, Pre-training Data=IG-3.6B, Evaluation Protocol=Linear Probing2022.12-7.7—18—80.4619.74
SWAGBackbone=ViT-B/16, Evaluation Protocol=Linear Probing, Pre-training Dataset=IG-3.6B2022.12-7.76——————
Uniform SoupArchitecture=ViT-B/32, (Pre)training data=WIT, Evaluation protocol=Fine-tuning2022.12-7.90.322423.87———
Uniform SoupBackbone=ViT-B/32, Evaluation Protocol=Fine-tuned, Pre-training Dataset=WIT2022.12-7.9——————
Uniform SoupBackbone=ViT-B/32, Pre-training Data=WIT, Evaluation Protocol=Fine-tuning2022.12-7.9—24—90.3223.87
BiT-MBackbone=ResNet-50v2, Pre-training Data=IN-21k2022.12-8.6—28—90.629.73
SupervisedArchitecture=ViT-B/32, (Pre)training data=IN-1k2022.12-8.71.23434.31———
SupervisedBackbone=ViT-B/32, Pre-training Data=IN-1k2022.12-8.7—34—91.234.31
ERMBackbone=ViT-B/322022.12-8.71——————
SupervisedArchitecture=RG-32gf, (Pre)training data=IN-1k2022.12-14.13.743233.43———
SupervisedBackbone=RG-32gf, Pre-training Data=IN-1k2022.12-14.1—32—93.7433.43
ERMBackbone=RG-32gf2022.12-14.15——————
CutMixArchitecture=ResNet-50, (Pre)training data=IN-1k2022.12-14.81.922229.61———
CutMixBackbone=ResNet-50, Pre-training Data=IN-1k2022.12-14.8—22—91.9229.61
MoCov3Backbone=ViT-B/16, Evaluation Protocol=Linear Probing2022.12-16——————
MoCov3Backbone=ViT-B/16, Pre-training Data=IN-1k, Evaluation Protocol=Linear Probing2022.12-16—22—91.9738.34
AugMixArchitecture=ResNet-50, (Pre)training data=IN-1k2022.12-16.82.613634.44———
AugMixBackbone=ResNet-50, Pre-training Data=IN-1k2022.12-16.8—36—92.6134.44
Style TransferArchitecture=ResNet-50, (Pre)training data=SIN2022.12-17.34.915250.06———
Style TransferBackbone=ResNet-50, Pre-training Data=SIN2022.12-17.3—52—104.9150.06
CutoutArchitecture=ResNet-50, (Pre)training data=IN-1k2022.12-182.933238.06———
CutoutBackbone=ResNet-50, Pre-training Data=IN-1k2022.12-18—32—82.9338.06
MixupArchitecture=ResNet-50, (Pre)training data=IN-1k2022.12-18.63.433839.78———
MixupBackbone=ResNet-50, Pre-training Data=IN-1k2022.12-18.6—38—73.4339.78
MoCov3Architecture=ResNet-50, (Pre)training data=IN-1k, Evaluation protocol=Linear Probing2022.12-20.72.944444.37———
MoCov3Backbone=ResNet-50, Pre-training Data=IN-1k, Evaluation Protocol=Linear Probing2022.12-20.7—44—82.9444.37
SupervisedArchitecture=ResNet-50, (Pre)training data=IN-1k2022.12-26.77.564042.46———
SupervisedBackbone=ResNet-50, Pre-training Data=IN-1k2022.12-26.7—40—77.5642.46