List your company
ResearchFormerly WizWand

Image Classification on Tiny-ImageNet (Top-1/Top-5 Accuracy)

90Top-1 AccuracyAdapterTune
53.017662.618872.2281.8212Nov 30, 2021Aug 18, 2022May 6, 2023Jan 22, 2024Oct 9, 2024Jun 27, 2025Mar 16, 2026
Updated 7mo ago

Evaluation results

MethodLinks
AdapterTuneBackbone=ViT-B/162026.0390——
AdapterTuneBackbone=ViT-S/162026.0384.95——
Head-OnlyBackbone=ViT-B/162026.0381.45——
AdapterTuneBackbone=DeiT-T2026.0376.35——
Head-OnlyBackbone=ViT-S/162026.0373.69——
Haar Wavelet (HWT, 3-path)Evaluation Protocol=10-Crop, Params (M)=20.78 (-12.4%)2026.0373.2491.481.02
SAMixBackbone=ResNeXt-502021.1172.18——
Full FTBackbone=ViT-B/162026.0371.99——
Full FTBackbone=ViT-S/162026.0371.38——
Haar Wavelet (HWT, 3-path)Evaluation Protocol=Single-Crop, Params (M)=20.78 (-12.4%)2026.0370.8490.151.175
AutoMixBackbone=ResNeXt-502021.1170.72——
Head-OnlyBackbone=DeiT-T2026.0370.28——
Hadamard Transform (HT, 3-path)Evaluation Protocol=10-Crop, Params (M)=20.78 (-12.4%)2026.0369.0689.291.228
Full FTBackbone=DeiT-T2026.0369.04——
SAMixBackbone=ResNet-182021.1168.89——
AutoMixBackbone=ResNet-182021.1167.33——
ManifoldMixBackbone=ResNeXt-502021.1167.3——
PuzzleMixBackbone=ResNeXt-502021.1166.92——
Hadamard Transform (HT, 3-path)Evaluation Protocol=Single-Crop, Params (M)=20.78 (-12.4%)2026.0366.6587.431.39
SaliencyMixBackbone=ResNeXt-502021.1166.55——
CutMixBackbone=ResNeXt-502021.1166.47——
MixupBackbone=ResNeXt-502021.1166.36——
ResizeMixBackbone=ResNeXt-502021.1165.87——
PuzzleMixBackbone=ResNet-182021.1165.81——
ResNet (our trial, baseline)Evaluation Protocol=10-Crop, Params (M)=23.722026.0365.6887.831.288
CutMixBackbone=ResNet-182021.1165.53——
FMixBackbone=ResNeXt-502021.1165.08——
VanillaBackbone=ResNeXt-502021.1165.04——
SaliencyMixBackbone=ResNet-182021.1164.6——
ManifoldMixBackbone=ResNet-182021.1164.15——
MixupBackbone=ResNet-182021.1163.86——
ResizeMixBackbone=ResNet-182021.1163.74——
FMixBackbone=ResNet-182021.1163.47——
ResNet (our trial, baseline)Evaluation Protocol=Single-Crop, Params (M)=23.722026.0363.2886.371.409
VanillaBackbone=ResNet-182021.1161.68——
TeacherTeacher Architecture=ResNet110, Student Architecture=ResNet32, Data Size=50K~100K2022.0560.71——
TeacherTeacher Architecture=ResNet110, Student Architecture=VGG11, Data Size=50K~100K2022.0560.71——
TeacherTeacher Architecture=ResNet110, Student Architecture=MobileNet, Data Size=50K~100K2022.0560.71——
EL2NBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.12026.0360.33——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.32026.0360.31——
MLTeacher Architecture=ResNet110, Student Architecture=MobileNet, Data Size=50K~100K2022.0560.07——
TeacherTeacher Architecture=VGG13, Student Architecture=VGG11, Data Size=50K~100K2022.0559.89——
Full DBBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.02026.0359.85——
FedCSBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.12026.0359.79——
GradNDBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.12026.0359.74——
GradNDBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.32026.0359.49——
DKDTeacher Architecture=ResNet110, Student Architecture=MobileNet, Data Size=50K~100K2022.0559.43——
FedCoreBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.12026.0359.34——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.52026.0359.27——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.12026.0359.23——
EL2NBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.32026.0359.22——
EL2NBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.52026.0359.13——
GradNDBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.52026.0358.88——
ForgettingBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.32026.0358.87——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.72026.0358.87——
FedCSBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.52026.0358.84——
FedCSBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.32026.0358.81——
ForgettingBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.72026.0358.8——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.92026.0358.78——
RandomBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.32026.0358.68——
FedCoreBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.92026.0358.64——
GradNDBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.72026.0358.61——
FedCoreBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.32026.0358.57——
RandomBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.12026.0358.42——
EL2NBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.72026.0358.36——
GradNDBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.92026.0358.33——
FedCSBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.72026.0358.32——
FedCSBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.92026.0358.25——
EL2NBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.92026.0358.18——
ForgettingBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.12026.0358.17——
ForgettingBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.52026.0358.15——
FedCoreBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.52026.0358.13——
KDTeacher Architecture=ResNet110, Student Architecture=MobileNet, Data Size=50K~100K2022.0557.85——
FedCoreBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.72026.0357.64——
MLTeacher Architecture=VGG13, Student Architecture=VGG11, Data Size=50K~100K2022.0557.46——
RandomBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.52026.0357.36——
MLTeacher Architecture=ResNet110, Student Architecture=VGG11, Data Size=50K~100K2022.0556.78——
RandomBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.72026.0356.7——
ForgettingBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.92026.0356.57——
MLTeacher Architecture=ResNet110, Student Architecture=ResNet32, Data Size=50K~100K2022.0556.56——
DKDTeacher Architecture=ResNet110, Student Architecture=VGG11, Data Size=50K~100K2022.0556.52——
StudentTeacher Architecture=ResNet110, Student Architecture=MobileNet, Data Size=50K~100K2022.0556.07——
DKDTeacher Architecture=ResNet110, Student Architecture=ResNet32, Data Size=50K~100K2022.0555.99——
DKDTeacher Architecture=VGG13, Student Architecture=VGG11, Data Size=50K~100K2022.0555.88——
RandomBackbone=ResNet-50, Imbalance Ratio (IR)=2, Label partitioning (pl)=0.1, Non-IID data distribution (α)=1.0, Pruning rate (pf)=0.92026.0355.67——
StudentTeacher Architecture=ResNet110, Student Architecture=ResNet32, Data Size=50K~100K2022.0555.47——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.92026.0355.38——
MEKD (soft)Teacher Architecture=ResNet110, Student Architecture=MobileNet, Data Size=10K2022.0554.93——
MEKD (hard)Teacher Architecture=ResNet110, Student Architecture=MobileNet, Data Size=10K2022.0554.71——
AFLSetting=alpha = 0.12024.0554.67——
AFLSetting=alpha = 0.012024.0554.67——
AFLSetting=s = 102024.0554.67——
AFLSetting=s = 52024.0554.67——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.12026.0354.65——
ForgettingBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.12026.0354.63——
GradNDBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.12026.0354.63——
FedCSBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.12026.0354.6——
RandomBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.12026.0354.52——
SCOPEBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.72026.0354.49——
EL2NBackbone=ResNet-50, Imbalance Ratio (IR)=5, Label partitioning (pl)=0.1, Non-IID data distribution (α)=0.1, Pruning rate (pf)=0.32026.0354.44——