List your company
ResearchFormerly WizWand

Image Classification on ImageNet-Adversarial (val)

75.8Top-1 AccuracyALIGN
4.24822.82441.459.976Nov 14, 2022Feb 15, 2023May 20, 2023Aug 21, 2023Nov 23, 2023Feb 24, 2024May 28, 2024
Updated 8mo ago

Evaluation results

MethodLinks
ALIGNzero-shot evaluation protocol=True2022.1175.8
ZEROBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-142024.0575.15
RLCFBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-14, Tuning=full image encoder tuning (Theta_i), Adaptation steps (t)=32024.0573.71
EVA CLIP-gzero-shot evaluation protocol=True2022.1173.6
OpenAI CLIP-Lzero-shot evaluation protocol=True2022.1170.8
Zero-ShotBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-142024.0568.82
RLCFBackbone=CLIP-ViT-B-16 + CLIP-ViT-L-14, Tuning=prompt tuning (Theta_etx), Adaptation steps (t)=12024.0565.45
ZEROBackbone=MaPLe2024.0563.32
ZEROBackbone=CLIP-ViT-B-16, Ensemble variant=true2024.0562.75
ZEROBackbone=CLIP-ViT-B-162024.0559.61
PromptAlignBackbone=MaPLe2024.0559.37
Open CLIP-Hzero-shot evaluation protocol=True2022.1159.3
TPTBackbone=MaPLe2024.0558.08
Open CLIP-gzero-shot evaluation protocol=True2022.1157.2
TPTBackbone=CLIP-ViT-B-162024.0554.77
Zero-ShotBackbone=MaPLe2024.0550.9
EnsembleBackbone=CLIP-ViT-B-162024.0549.89
Zero-ShotBackbone=CLIP-ViT-B-162024.0547.87
TinyMIMModel Size=ViT-B2023.0143
MAEModel Size=ViT-B2023.0133.6
TinyMIMModel Size=ViT-S2023.0127.5
DeiTModel Size=ViT-B2023.0125.8
MAEModel Size=ViT-S2023.0120.1
DeiTModel Size=ViT-S2023.0118.3
TinyMIMModel Size=ViT-T2023.0111
DeiTModel Size=ViT-T2023.018
MAEModel Size=ViT-T2023.017