List your company
ResearchFormerly WizWand

Image Classification on STL10 (Accuracy)

99.6AccuracyBASIC-L
69.85677.57885.393.022Mar 17, 2021Jan 25, 2022Dec 5, 2022Oct 16, 2023Aug 25, 2024Jul 5, 2025May 16, 2026
Updated 4mo ago

Evaluation results

MethodLinks
BASIC-LBackbone=BASIC-L2021.1199.6
CLIPBackbone=ViT-L/14-3362021.1199.4
BASIC-MBackbone=BASIC-M2021.1199.3
CLIP (reported)Pre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.0799.3
CLIPPre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.0799.3
DaVinciLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0199.2
X-FM_baseLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0199.2
MLCDPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.0799.1
MLCDPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.0799.1
CLIPLinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0199
FLAVALinear evaluation=true, Model size=Base, Patch size=16*16, Resolution=224*2242023.0198.9
ITOBackbone=ViT-L/16, Pre-training Dataset=DataComp-1B, Training Epochs=1, Evaluation Protocol=Zero-shot2026.0398.6
FLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.0798.5
CLIPModel Type=Teacher, Img. Enc.=ViT-B/16, Param. Size=76M, zero-shot=true2023.0198.2
CLIPBackbone=ViT-B/162021.1198.2
CLIPPre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.0798.2
OpenCLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Zero-shot classification2024.0798.2
ITO sub2Backbone=ViT-L/16, Pre-training Dataset=DataComp-1B, Training Epochs=1, Evaluation Protocol=Zero-shot2026.0398.2
CLIPBackbone=ViT-L/16, Pre-training Dataset=DataComp-1B, Training Epochs=1, Evaluation Protocol=Zero-shot2026.0397.5
BeamCLIPModel Type=Student, Img. Enc.=RN50, Param. Size=24M2023.0197.45
OpenCLIPPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.0797.1
ITOZero-shot=true, Backbone=ViT-B/16, Pre-training Dataset=Laion100M, Epochs=302026.0397
Simple AveragingMerging Strategy=Simple Averaging, Quantization Method=Full-precision, Bit-width=Full, Backbone=CLIP-ViT-B/322026.0597
CLIPBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.0396.89
BASIC-SBackbone=BASIC-S2021.1196.7
CLIP+Pre-training Data=WIT-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.0796.7
UNICOMPre-training Data=LAION-400M, Backbone=ViT-L/14, Evaluation Protocol=Linear Probe2024.0796.7
Simple Averaging + GPTQMerging Strategy=Simple Averaging, Quantization Method=GPTQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0596.53
Simple Averaging + RTNMerging Strategy=Simple Averaging, Quantization Method=RTN, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0596.35
WUDI-Merging + E-PMQMerging Strategy=WUDI-Merging, Quantization Method=E-PMQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0596.3
EnsemblingBackbone=ViT-B/322026.0296
Simple Averaging + AWQMerging Strategy=Simple Averaging, Quantization Method=AWQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0595.94
AGFTBackbone=ViT-B/32, Evaluation Protocol=Zero-shot, Training Strategy=Adversarially fine-tuned2026.0395.72
GLADIATORBackbone=ViT-B/32, Evaluation Protocol=Zero-shot, Training Strategy=Adversarially fine-tuned2026.0395.71
CLIPZero-shot=true, Backbone=ViT-B/16, Pre-training Dataset=Laion100M, Epochs=302026.0395.5
Simple Averaging + E-PMQMerging Strategy=Simple Averaging, Quantization Method=E-PMQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0595.25
TIES-Merging + E-PMQMerging Strategy=TIES-Merging, Quantization Method=E-PMQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0595.1
nCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.1094.8
ReMixMatchLearning paradigm=Semi-supervised2021.0394.8
PMG-AFTBackbone=ViT-B/32, Evaluation Protocol=Zero-shot, Training Strategy=Adversarially fine-tuned2026.0394.73
CLIPBackbone=ResNet-502021.1194.3
SLIPZero-shot=true, Backbone=ViT-B/16, Pre-training Dataset=Laion100M, Epochs=302026.0393.8
TGA-ZSRBackbone=ViT-B/32, Evaluation Protocol=Zero-shot, Training Strategy=Adversarially fine-tuned2026.0393.52
Task Arithmetic + E-PMQMerging Strategy=Task Arithmetic, Quantization Method=E-PMQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0593.44
WUDI-MergingMerging Strategy=WUDI-Merging, Quantization Method=Full-precision, Bit-width=Full, Backbone=CLIP-ViT-B/322026.0593.4
KDModel Type=Student, Img. Enc.=RN50, Param. Size=24M2023.0193.28
TeCoABackbone=ViT-B/32, Evaluation Protocol=Zero-shot, Training Strategy=Adversarially fine-tuned2026.0393.23
SPICELearning paradigm=Unsupervised2021.0392.9
xCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.1092.8
WUDI-Merging + GPTQMerging Strategy=WUDI-Merging, Quantization Method=GPTQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0592.75
UDALearning paradigm=Semi-supervised2021.0392.3
WUDI-Merging + RTNMerging Strategy=WUDI-Merging, Quantization Method=RTN, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0592.2
WUDI-Merging + AWQMerging Strategy=WUDI-Merging, Quantization Method=AWQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0592.01
FixMatchLearning paradigm=Semi-supervised2021.0392
CLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.1091.3
DLMEEvaluation Protocol=linear-test2022.0790.1
CyCLIPprotocol=Linear probing, backbone=ResNet-502022.0590.1
TIES-Merging + GPTQMerging Strategy=TIES-Merging, Quantization Method=GPTQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0589.93
M-TIESBackbone=ViT-B/322026.0289.9
TIES-MergingMerging Strategy=TIES-Merging, Quantization Method=Full-precision, Bit-width=Full, Backbone=CLIP-ViT-B/322026.0589.9
MixMatchLearning paradigm=Semi-supervised2021.0389.6
TIES-Merging + RTNMerging Strategy=TIES-Merging, Quantization Method=RTN, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0589.53
TIES-Merging + AWQMerging Strategy=TIES-Merging, Quantization Method=AWQ, Bit-width=4-bit, Backbone=CLIP-ViT-B/322026.0589.35
CLIPprotocol=Linear probing, backbone=ResNet-502022.0589.23
Deep Laplacian eigenmapsEvaluation protocol=linear evaluation protocol2022.1088.97
BYOLEvaluation Protocol=linear-test2022.0788.7
Barlow TwinsEvaluation protocol=linear evaluation protocol2022.1088.65
BYOLEvaluation protocol=linear evaluation protocol2022.1088.46
DLME-A1Evaluation Protocol=linear-test2022.0788.1
SLIPPre-train dataset=CC3M, Backbone=ViT-B/16, Training Epochs=30, Evaluation Protocol=Zero-shot2026.0388.1
ITO sub2Pre-train dataset=CC3M, Backbone=ViT-B/16, Training Epochs=30, Evaluation Protocol=Zero-shot2026.0388.1
DLME-A2Evaluation Protocol=linear-test2022.0787.9
DAREBackbone=ViT-B/322026.0287.8
MoCo v2Evaluation protocol=linear evaluation protocol2022.1087.71
TIESBackbone=ViT-B/322026.0287.6
SimCLREvaluation protocol=linear evaluation protocol2022.1087.52
DLME-A3Evaluation Protocol=linear-test2022.0787.3
SimCLREvaluation Protocol=linear-test2022.0786.9
MoCo.v2Evaluation Protocol=linear-test2022.0785.6
ITOPre-train dataset=CC3M, Backbone=ViT-B/16, Training Epochs=30, Evaluation Protocol=Zero-shot2026.0385.6
InvL-DNPδ (noise level)=0.07, Backbone=ResNet, Batch size=10, FC layer size=202025.1284
DNPδ (noise level)=0.03, Backbone=ResNet, Batch size=10, FC layer size=202025.1283.3
FLAIRPre-train dataset=CC3M, Backbone=ViT-B/16, Training Epochs=30, Evaluation Protocol=Zero-shot2026.0383.1
NPIDEvaluation Protocol=linear-test2022.0782.5
DINORANKCLIPProtocol=Linear probing2026.0582.4
CyCLIPProtocol=Linear probing2026.0582.3
CRDModel Type=Student, Img. Enc.=RN50, Param. Size=24M2023.0181.41
InvL-GNPδ (noise level)=0.005, Backbone=ResNet, Batch size=10, FC layer size=202025.1281.3
ALIPProtocol=Linear probing2026.0581
GNPδ (noise level)=0.001, Backbone=ResNet, Batch size=10, FC layer size=202025.1280.2
RANKCLIPProtocol=Linear probing2026.0579.6
CLIPProtocol=Linear probing2026.0579
Mean TeacherLearning paradigm=Semi-supervised2021.0378.6
SCANLearning paradigm=Unsupervised2021.0376.7
Π-ModelLearning paradigm=Semi-supervised2021.0374.8
ODCEvaluation Protocol=linear-test2022.0773.4
H-SCL(β)Evaluation Protocol=Linear probing, Backbone=ResNet-50, Projection head dimension=128, Batch size=512, Epochs=2002022.0872.52
Pseudo-LabelingLearning paradigm=Semi-supervised2021.0372
H-SCL(τ)Evaluation Protocol=Linear probing, Backbone=ResNet-50, Projection head dimension=128, Batch size=512, Epochs=2002022.0871.02
Task ArithmeticMerging Strategy=Task Arithmetic, Quantization Method=Full-precision, Bit-width=Full, Backbone=CLIP-ViT-B/322026.0571