List your company
ResearchFormerly WizWand

Image Classification on ImageNet100-B (test)

8,990ASRCorruptEncoder
-359.62,067.74,4956,922.3Nov 15, 2022
Updated 8mo ago

Evaluation results

MethodLinks
CorruptEncoderBackbone=ResNet-18, CL algorithm=MoCo-v2, Evaluation Protocol=linear evaluation2022.118,990——
Poisoned-EncoderBackbone=ResNet-18, CL algorithm=MoCo-v2, Evaluation Protocol=linear evaluation2022.115,320——
CTRLBackbone=ResNet-18, CL algorithm=MoCo-v2, Evaluation Protocol=linear evaluation2022.112,050——
SSL-BackdoorBackbone=ResNet-18, CL algorithm=MoCo-v2, Evaluation Protocol=linear evaluation2022.111,430——
CorruptEncoderTarget Class=Street Sign, Pre-training dataset=Conceptual Captions [27]2022.1197.7—49
CorruptEncoderTarget Class=Ski Mask, Pre-training dataset=Conceptual Captions [27]2022.1196.6—48.6
Carlini and TerzisTarget Class=Ski Mask, Pre-training dataset=Conceptual Captions [27]2022.1196—48.5
Carlini and TerzisTarget Class=Street Sign, Pre-training dataset=Conceptual Captions [27]2022.1194—48.3
CorruptEncoderTarget Class=Lorikeet, Pre-training dataset=Conceptual Captions [27]2022.1189—48.4
CorruptEncoderTarget Class=Komondor, Pre-training dataset=Conceptual Captions [27]2022.1160.9—48.8
CorruptEncoderTarget Class=Rottweiler, Pre-training dataset=Conceptual Captions [27]2022.1157—48.9
No AttackBackbone=ResNet-18, CL algorithm=MoCo-v2, Evaluation Protocol=linear evaluation2022.1140——
No AttackTarget Class=Lorikeet, Pre-training dataset=Conceptual Captions [27]2022.111.948.4—
No AttackTarget Class=Rottweiler, Pre-training dataset=Conceptual Captions [27]2022.111.748.4—
No AttackTarget Class=Ski Mask, Pre-training dataset=Conceptual Captions [27]2022.111.448.4—
No AttackTarget Class=Street Sign, Pre-training dataset=Conceptual Captions [27]2022.11148.4—
No AttackTarget Class=Komondor, Pre-training dataset=Conceptual Captions [27]2022.110.348.4—
Carlini and TerzisTarget Class=Lorikeet, Pre-training dataset=Conceptual Captions [27]2022.110.1—47.7
Carlini and TerzisTarget Class=Rottweiler, Pre-training dataset=Conceptual Captions [27]2022.110—48.6
Carlini and TerzisTarget Class=Komondor, Pre-training dataset=Conceptual Captions [27]2022.110—48.9