List your company
ResearchFormerly WizWand

Image Classification on TinyImageNet

92AccuracyOCD
33.926449.003264.0879.1568May 18, 2022Jan 15, 2023Sep 14, 2023May 14, 2024Jan 11, 2025Sep 10, 2025May 11, 2026
Updated 5mo ago

Evaluation results

MethodLinks
OCDArchitecture=DeiT-B/16-D, Ensemble count=52022.1092—0.65
Swin L/4Architecture=Swin L/42022.1091.35——
OCDArchitecture=DeiT-B/16-D, Ensemble count=12022.1090.8—0.65
TentArchitecture=DeiT-B/16-D, Number of iterations=102022.1088.2—0.7
TentArchitecture=DeiT-B/16-D, Number of iterations=302022.1088.2—0.07
TentArchitecture=DeiT-B/16-D, Number of iterations=12022.1088.15—0.71
DeiT-B/16-DArchitecture=DeiT-B/16-D2022.1087.29—0.71
WaveMix-192/16 (level 3)Input Resolution=128 x 128, Pre-trained on ImageNet-1k=true, Augmentation=TrivialAugment [78]2022.0577.49——
B-12/Wi-1024 + DA + TTABlocks=12, Width=1024, Data Augmentation=true, Test-Time Augmentation=true, Pre-training=ImageNet21k, Evaluation Protocol=Fine-tuning2023.0673.1——
LeRaCModel=CvT-13 (pre-trained), Training Regime=LeRaC (ours)2022.0571.34——
conventionalModel=CvT-13 (pre-trained), Training Regime=conventional2022.0570.71——
B-12/Wi-1024 + DABlocks=12, Width=1024, Data Augmentation=true, Test-Time Augmentation=false, Pre-training=ImageNet21k, Evaluation Protocol=Fine-tuning2023.0669.9——
CBSModel=CvT-13 (pre-trained), Training Regime=CBS2022.0568.41——
TeacherTeacher Architecture=ResNet-34, Student Architecture=ResNet-182023.0966.44——
ResNet-18Params(M)=11.7, FLOPs(M)=1800, Notes=Verified across re-implementations2025.1265——
KvLIFNetwork Architecture=VGG-13, Timestep=62026.0364.82——
CPT-SNNNetwork Architecture=VGG-13, Timestep=82026.0364.78——
WaveMix-244/12 (level 2)Input Resolution=64 x 64, Pre-trained on ImageNet-1k=false, Augmentation=TrivialAugment [78]2022.0564.51——
NAYERTeacher Architecture=ResNet-34, Student Architecture=ResNet-18, E=3002023.0964.17——
CLIFNetwork Architecture=VGG-13, Timestep=62026.0364.13——
SpaceshipNetTeacher Architecture=ResNet-34, Student Architecture=ResNet-182023.0964.04——
KvLIFNetwork Architecture=VGG-13, Timestep=42026.0364.03——
WaveMix-Lite/WaveMix-32Params(M)=1–2, FLOPs(M)=250, Notes=Reported in WaveMix paper2025.1264——
CPT-SNNNetwork Architecture=VGG-13, Timestep=42026.0363.39——
CLIFNetwork Architecture=VGG-13, Timestep=42026.0363.16——
B-6/Wi-1024 + DABlocks=6, Width=1024, Data Augmentation=true, Test-Time Augmentation=false, Pre-training=ImageNet21k, Evaluation Protocol=Fine-tuning2023.0662.7——
MD-SNNModel=ResNet34, Precision (W / U)=8 / 8, Timestep=42025.1260.59——
MixNet-SParams(M)=4.1, FLOPs(M)=256, Notes=timm community runs2025.1260——
MD-SNNModel=ResNet34, Precision (W / U)=4 / 4, Timestep=42025.1259.9——
TinyNet-AParams(M)=6.2, FLOPs(M)=339, Notes=timm community runs2025.1259——
KD+ATS+EnsTeacher=WRN50-2, Student=MV22022.1058.4763.54—
MobileNetV2Params(M)=2.2, FLOPs(M)=330, Notes=Common open-source benchmarks2025.1258——
LeRaCModel=ResNet-18, Training Regime=LeRaC (ours)2022.0557.86——
conventionalModel=ResNet-18, Training Regime=conventional2022.0557.41——
ResKDTeacher=WRN50-2, Student=MV22022.1057.3262.26—
ASGLNetwork Architecture=VGG-13, Timestep=82026.0356.81——
ASGLNetwork Architecture=VGG-13, Timestep=42026.0356.57——
LeRaCModel=Wide ResNet-50, Training Regime=LeRaC (ours)2022.0556.48——
EnsTeacher=WRN50-2, Student=MV22022.1056.461.82—
conventionalModel=Wide ResNet-50, Training Regime=conventional2022.0555.97——
CBSModel=ResNet-18, Training Regime=CBS2022.0555.49——
Joint A-SNNNetwork Architecture=VGG-16, Timestep=42026.0355.39——
ShuffleNet V2 (0.5x)Params(M)=1.26, FLOPs(M)=150, Notes=Public leaderboard reports2025.1255——
KD+ATSTeacher=WRN50-2, Student=MV22022.1054.1160.88—
SpikeSimModel=VGG16, Precision (W / U)=8 / 32, Timestep=102025.1254.03——
ST-KDTeacher=WRN50-2, Student=MV22022.1052.9359.77—
TAKDTeacher=WRN50-2, Student=MV22022.1052.4458.87—
KDTeacher=WRN50-2, Student=MV22022.1052.3358.89—
ESKDTeacher=WRN50-2, Student=MV22022.1052.1559.01—
KD+ATS+EnsTeacher=WRN50-2, Student=SFV22022.1052.1463.54—
NoKDTeacher=WRN50-2, Student=MV22022.1052.0357.94—
ResKDTeacher=WRN50-2, Student=SFV22022.1051.9362.26—
Mixed Barlow TwinsBackbone=ResNet-50, Epochs=2000, Evaluation Protocol=linear2023.1251.84——
SCKDTeacher=WRN50-2, Student=MV22022.1051.8359.04—
Mixed Barlow TwinsBackbone=ResNet-18, Epochs=2000, linear=true2023.1251.67——
Mixed Barlow TwinsBackbone=ResNet-18, Epochs=1000, linear=true2023.1251.48——
BYOLBackbone=ResNet-18, Epochs=1000, linear=true2023.1251——
EnsTeacher=WRN50-2, Student=SFV22022.1050.6961.82—
Mixed Barlow TwinsBackbone=ResNet-50, Epochs=1000, Evaluation Protocol=linear2023.1250.59——
KD+ATSTeacher=WRN50-2, Student=SFV22022.1050.0360.88—
ST-KDTeacher=WRN50-2, Student=SFV22022.1049.5959.77—
SimCLRBackbone=ResNet-18, Epochs=1000, linear=true2023.1248.84——
SCKDTeacher=WRN50-2, Student=SFV22022.1048.7659.04—
TAKDTeacher=WRN50-2, Student=SFV22022.1048.7158.87—
KDTeacher=WRN50-2, Student=SFV22022.1048.4858.89—
ESKDTeacher=WRN50-2, Student=SFV22022.1048.3459.01—
CBSModel=Wide ResNet-50, Training Regime=CBS2022.0548.3——
W-MSEBackbone=ResNet-18, Epochs=1000, linear=true2023.1248.2——
B-6/Wi-1024Blocks=6, Width=1024, Data Augmentation=false, Test-Time Augmentation=false, Pre-training=ImageNet21k, Evaluation Protocol=Fine-tuning2023.0647.1——
GlimmerNetParams(M)=0.24, FLOPs(M)=28, Notes=Scaled-up configuration2025.1247.08——
COSMOSalpha=5, heterogeneity=Dirichlet, round=final2026.0546.9——
Barlow TwinsBackbone=ResNet-50, Epochs=1000, Evaluation Protocol=linear2023.1246.86——
SimCLRBackbone=ResNet-50, Epochs=1000, Evaluation Protocol=linear2023.1246.34——
COSMOSdata heterogeneity=Dirichlet, alpha=100, performance type=Final-round2026.0546.2——
NoKDTeacher=WRN50-2, Student=SFV22022.1045.7957.94—
MINTModel=ResNet34, Precision (W / U)=4 / 4, Timestep=42025.1244.94——
MINTModel=ResNet34, Precision (W / U)=8 / 8, Timestep=42025.1242.94——
Mixed Barlow TwinsBackbone=ResNet-50, Epochs=2000, Evaluation Protocol=k-NN2023.1242.21——
Barlow TwinsBackbone=ResNet-18, Epochs=1000, linear=true2023.1241.8——
MoCoBackbone=ResNet-18, Evaluation Protocol=Linear evaluation2025.0941.23——
BYOLBackbone=ResNet-50, Epochs=1000, Evaluation Protocol=linear2023.1241.02——
Mixed Barlow TwinsBackbone=ResNet-50, Epochs=1000, Evaluation Protocol=k-NN2023.1240.52——
KD+ATS+EnsTeacher=WRN50-2, Student=ANet2022.1040.4263.54—
Kernel VICReg—RQBackbone=ResNet-18, Evaluation Protocol=Linear evaluation, Kernel=RQ2025.0940.38——
Kernel VICReg—LaplacianBackbone=ResNet-18, Evaluation Protocol=Linear evaluation, Kernel=Laplacian2025.0940.12——
EnsTeacher=WRN50-2, Student=ANet2022.1039.3761.82—
ResKDTeacher=WRN50-2, Student=ANet2022.1038.6662.26—
Kernel VICReg—RBFBackbone=ResNet-18, Evaluation Protocol=Linear evaluation, Kernel=RBF2025.0938.21——
Mixed Barlow TwinsBackbone=ResNet-18, Epochs=2000, k-NN=true2023.1238.11——
SimCLRBackbone=ResNet-18, Evaluation Protocol=Linear evaluation2025.0937.83——
Barlow TwinsBackbone=ResNet-50, Epochs=1000, Evaluation Protocol=k-NN2023.1237.66——
Whole DatasetImg/Cls=N/A, Ratio (%)=1002023.0737.6——
Mixed Barlow TwinsBackbone=ResNet-18, Epochs=1000, k-NN=true2023.1237.52——
Kernel VICReg—LinearBackbone=ResNet-18, Evaluation Protocol=Linear evaluation, Kernel=Linear2025.0937.48——
KD+ATSTeacher=WRN50-2, Student=ANet2022.1037.4260.88—
COMETdata heterogeneity=Dirichlet, alpha=100, performance type=Final-round2026.0537.1——
BYOLBackbone=ResNet-18, Evaluation Protocol=Linear evaluation2025.0936.31——
BYOLBackbone=ResNet-18, Epochs=1000, k-NN=true2023.1236.24——
TAKDTeacher=WRN50-2, Student=ANet2022.1036.258.87—
ST-KDTeacher=WRN50-2, Student=ANet2022.1036.1659.77—