List your company
ResearchFormerly WizWand

Image Classification on ImageNet-Adversarial

87.3Top-1 AccEVA-CLIP-18B
33.4847.452561.42575.3975Dec 1, 2022Jul 1, 2023Jan 29, 2024Aug 29, 2024Mar 29, 2025Oct 27, 2025May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
EVA-CLIP-18BZero-shot=true2024.0287.3
EVA-CLIP-8BZero-shot=true2024.0285.2
InternVL-CZero-shot=true2024.0283.8
EVA-02-CLIP-E/14+zero-shot evaluation protocol=true2023.0382.1
EVA-02-CLIP-E/14+Zero-shot=true2024.0282.1
DFN5B-CLIP-H/14+Zero-shot=true2024.0279.6
CLIPModel=ViT-L/14@336, Pre-training Data=WIT-400M, Evaluation Protocol=our eval., Image size=3362022.1278
CLIPModel=ViT-L/14@336, Pre-training Data=WIT-400M, Image size=3362022.1277.2
EVA-01-CLIP-g/14+Zero-shot=true2024.0274.1
EVA-01-CLIP-g/14Zero-shot=true2024.0273.6
CLIPModel=ViT-L/14, Pre-training Data=WIT-400M, Evaluation Protocol=our eval., Image size=2242022.1271.9
DFN5B-CLIP-H/14Zero-shot=true2024.0271.7
OpenCLIP-G/14Zero-shot=true2024.0269.3
TRACERBackbone=ViT-B/16, Pre-trained Model=CLIP, Mode=Finetuning2026.0554.92
EVA-02-CLIP-B/16zero-shot evaluation protocol=true2023.0354.1
MERGETUNE + Weight ens.Backbone=CLIP ViT-B/16, Protocol=E2E-FT2026.0152.68
MERGETUNEBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.0151.61
CaRotBackbone=ViT-B/16, Pre-trained Model=CLIP, Mode=Finetuning2026.0551.57
FLIPModel=ViT-L/14, Pre-training Data=LAION-400M, Image size=2242022.1251.2
Zero-shot (CLIP)Backbone=CLIP ViT-B/16, Protocol=Zero-shot2026.0150.12
MERGETUNE + Weight ens.Backbone=CLIP ViT-B/16, Protocol=Linear Probing2026.0150.04
OpenAI CLIP-B/16zero-shot evaluation protocol=true2023.0350
ZSBackbone=ViT-B/16, Pre-trained Model=CLIP, Mode=Zero-Shot2026.0549.95
Lipsum-FTBackbone=ViT-B/16, Pre-trained Model=CLIP, Mode=Finetuning2026.0549.87
LP-FTBackbone=ViT-B/16, Pre-trained Model=CLIP, Mode=Finetuning2026.0549.28
MERGETUNEBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.0149.21
FLYPBackbone=ViT-B/16, Pre-trained Model=CLIP, Mode=Finetuning2026.0548.49
VRFBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.0148.41
TIESBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.0148.32
OpenCLIPModel=ViT-L/14, Pre-training Data=LAION-400M, Evaluation Protocol=our eval., Image size=2242022.1248.3
VRFBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.0148.08
Weight ens.Backbone=CLIP ViT-B/16, Protocol=Linear Probing2026.0147.89
TIESBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.0147.81
DAREBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.0147.74
Weight ens.Backbone=CLIP ViT-B/16, Protocol=E2E-FT2026.0147.61
Linear ProbingBackbone=CLIP ViT-B/16, Protocol=Linear Probing2026.0146.48
CLIPModel=ViT-L/14, Pre-training Data=LAION-400M, Evaluation Protocol=our repro., Image size=2242022.1246.3
E2E-FTBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.0136.62
DAREBackbone=CLIP ViT-B/16, Protocol=E2E-FT2026.0135.55