List your company
ResearchFormerly WizWand

Image Classification on ImageNet (Top-1 Accuracy)

94.43Top-1 AccuracyCoCoOp
73.806879.160984.51589.8691Jun 18, 2020Apr 2, 2021Jan 15, 2022Oct 31, 2022Aug 15, 2023May 29, 2024Mar 14, 2025
Updated 7mo ago

Evaluation results

MethodLinks
CoCoOpZSL setting=Inductive, Backbone=ViT-B/162024.0394.43
CoOpZSL setting=Inductive, Backbone=ViT-B/162024.0393.7
PromptKDZSL setting=Transductive, Backbone=ViT-B/162024.0393.61
PromptSRCZSL setting=Inductive, Backbone=ViT-B/162024.0393.6
MaPLeZSL setting=Inductive, Backbone=ViT-B/162024.0393.53
M2-EncoderMode=Zero-shot, Parameters=10B2024.0188.5
CoCaEvaluation Protocol=Zero-shot2022.0586.3
CoCaZero-shot=true2022.0986.3
CoCaMode=Zero-shot, Parameters=2.1B2024.0186.3
BASICEvaluation Protocol=Zero-shot2022.0585.7
BASIC-LBackbone=BASIC-L2021.1185.7
BASICZero-shot=true2022.0985.7
X-FM baseEval=FT, Multi-modal pre-training data size=1.3B2023.0185.7
BASICMode=Zero-shot, Parameters=3B2024.0185.7
BEiTv2Eval=FT2023.0185.5
X-FM baseEval=FT, Multi-modal pre-training data size=4M2023.0185.5
LiT ViT-eZero-shot=true2022.0985.4
M2-EncoderMode=Zero-shot, Parameters=1B2024.0185.4
OmniVLEval=FT, Multi-modal pre-training data size=14M2023.0185.3
LiT ViT-gZero-shot=true2022.0985.2
CoCa-LargeEvaluation Protocol=Zero-shot2022.0584.8
CoCaMode=Zero-shot, Parameters=0.8B2024.0184.8
LiTEvaluation Protocol=Zero-shot2022.0584.5
OFAEval=FT, Multi-modal pre-training data size=21M2023.0184.5
LiTMode=Zero-shot, Parameters=1B2024.0184.5
SimVLMEval=FT, Multi-modal pre-training data size=1.8B2023.0183.9
FlorenceEvaluation Protocol=Zero-shot2022.0583.7
FlorenceMode=Zero-shot, Parameters=0.9B2024.0183.7
CoCa-BaseEvaluation Protocol=Zero-shot2022.0582.6
CoCaMode=Zero-shot, Parameters=0.4B2024.0182.6
X2-VLMEval=FT, Multi-modal pre-training data size=1.3B2023.0182.2
LeRaCModel=CvT-13, Training Regime=LeRaC (ours)2022.0582.19
X-FM baseEval=LE, Multi-modal pre-training data size=1.3B2023.0182
EVAMode=Zero-shot, Parameters=5.0B2024.0182
Dr. ViTBackbone=ViT-B, Resolution=384x3842021.1181.83
ViT-BBackbone=ViT-B, Resolution=384x3842021.1181.63
Distill on real imagesTeacher Network=ResNet-101, Teacher Accuracy=77.37%, Student Network=DeiT-S, Dataset Partition=ImageNet2022.0481.5
BASIC-MBackbone=BASIC-M2021.1181.5
BASICMode=Zero-shot, Parameters=0.4B2024.0181.5
conventionalModel=CvT-13, Training Regime=conventional2022.0581.33
X-FM_baseLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0181.2
X-FM baseEval=LE, Multi-modal pre-training data size=4M2023.0181.2
OmniVLEval=LE, Multi-modal pre-training data size=14M2023.0181
X2-VLMEval=FT, Multi-modal pre-training data size=4M2023.0180.8
X2-VLMEval=LE, Multi-modal pre-training data size=1.3B2023.0180.6
CBSModel=CvT-13, Training Regime=CBS2022.0580.42
CLIPLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0180.2
BEiTv2Eval=LE2023.0180.1
OpenCLIPMode=Zero-shot, Parameters=2.5B2024.0180
Train from scratchTeacher Network=ResNet-101, Teacher Accuracy=77.37%, Student Network=DeiT-S2022.0479.8
2SFS LoRABackbone=ViT-L/14, Shots (k=16)=16, Evaluation Protocol=all-to-all2025.0379.7
Dr. ViTBackbone=ViT-B, Discrete Only=false2021.1179.48
ViT-BBackbone=ViT-B2021.1178.73
Dr. ViTBackbone=ViT-B, Discrete Only=true2021.1178.67
OpenCLIPMode=Zero-shot, Parameters=1.3B2024.0178.5
EVAMode=Zero-shot, Parameters=1.1B2024.0178.5
M2-EncoderMode=Zero-shot, Parameters=0.4B2024.0178.5
FILIPEvaluation Protocol=Zero-shot2022.0578.3
FILIPMode=Zero-shot, Parameters=0.4B2024.0178.3
CDARTS-bParams(M)=6.42020.0678.2
DaVinciLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0177.7
Uni-PerEval=LE, Multi-modal pre-training data size=30M2023.0177.7
FairNAS-AParams(M)=5.92020.0677.5
CDARTS-aParams(M)=72020.0677.4
Dr. ViTBackbone=ViT-S, Discrete Only=false2021.1177.03
MixNet-MParams(M)=52020.0677
MixPath-AParams(M)=52020.0676.9
ResNet-50Backbone=ResNet-502021.1176.61
ALIGNEvaluation Protocol=Zero-shot2022.0576.4
ALIGNZero-shot=true2022.0976.4
ALIGNMode=Zero-shot, Parameters=0.8B2024.0176.4
CLIPEvaluation Protocol=Zero-shot2022.0576.2
CLIPBackbone=ViT-L/14-3362021.1176.2
CLIPZero-shot=true2022.0976.2
CLIPMode=Zero-shot, Parameters=0.4B2024.0176.2
OFAEval=LE, Multi-modal pre-training data size=21M2023.0175.9
AdCoEpochs=800, Multi-crop=true, Evaluation protocol=linear evaluation2021.1175.7
ExactOBSModel=ResNet50, FLOP reduction target=2x2022.0875.64
MnasNet-A2Params(M)=4.82020.0675.6
NNCLREpochs=1000, Multi-crop=true, Evaluation protocol=linear evaluation2021.1175.6
AdaPruneModel=ResNet50, FLOP reduction target=2x2022.0875.53
UniGrad (+ Cut-Mix)Epochs=800, Multi-crop=true, Evaluation protocol=linear evaluation2021.1175.5
FLAVALinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0175.5
X2-VLMEval=LE, Multi-modal pre-training data size=4M2023.0175.5
L-OBSModel=ResNet50, FLOP reduction target=2x2022.0875.48
SCEEpochs=200, Multi-crop=true, Evaluation protocol=linear evaluation2021.1175.4
SwaVEpochs=800, Multi-crop=true, Evaluation protocol=linear evaluation2021.1175.3
DINOEpochs=800, Multi-crop=true, Evaluation protocol=linear evaluation2021.1175.3
LlipData Size=2.5B, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.1275.3
ViT-SBackbone=ViT-S2021.1175.21
MobileNetV32020.0675.2
Hybrid ViT-SBackbone=Hybrid ViT-S2021.1175.1
ExactOBSModel=ResNet50, FLOP reduction target=3x2022.0875.01
SPOSParams(M)=4.32020.0675
Hybrid ViT-BBackbone=Hybrid ViT-B2021.1174.94
GMPModel=ResNet50, FLOP reduction target=2x2022.0874.86
ReSSLEpochs=200, Multi-crop=true, Evaluation protocol=linear evaluation2021.1174.7
WCLEpochs=800, Multi-crop=true, Evaluation protocol=linear evaluation2021.1174.7
Distill on real imagesTeacher Network=ResNet-101, Teacher Accuracy=77.37%, Student Network=DeiT-Ti, Dataset Partition=ImageNet2022.0474.6
ALABackbone=NASNet-A, length=2002023.1074.6