List your company
ResearchFormerly WizWand

Image Classification on Imagenette

98AccuracyUMAP
30.618448.111765.60583.0983Dec 6, 2021Sep 10, 2022Jun 16, 2023Mar 20, 2024Dec 24, 2024Sep 28, 2025Jul 4, 2026
Updated 3mo ago

Evaluation results

MethodLinks
UMAPSupervised Evaluator=5-NN classifier, Embedding=UMAP, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0398
Finsler UMAPSupervised Evaluator=5-NN classifier, Embedding=Finsler UMAP, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0398
t-SNESupervised Evaluator=5-NN classifier, Embedding=t-SNE, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0398
Finsler t-SNESupervised Evaluator=5-NN classifier, Embedding=Finsler t-SNE, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0398
Finsler UMAPSupervised Evaluator=Linear classifier, Embedding=Finsler UMAP, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0398
t-SNESupervised Evaluator=Linear classifier, Embedding=t-SNE, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0398
Finsler t-SNESupervised Evaluator=Linear classifier, Embedding=Finsler t-SNE, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0398
UMAPSupervised Evaluator=Linear classifier, Embedding=UMAP, Evaluation Protocol=2-stratified 5-fold CV (10 runs)2026.0397
RDCNet2026.0493.48
TeacherBackbone=ResNet182026.0491.29
TeacherResolution=high-resolution2026.0490.47
Student-FullResolution=high-resolution2026.0489.89
SSCNet2026.0488.75
QKFormer2026.0488.42
WRN-28-102026.0488.34
FAVOR+2026.0488.16
EfficientNet2026.0488.06
Couplformer2026.0487.91
GhostNet2026.0487.83
ResNet-342026.0487.72
CAPR-DenseNet2026.0487.72
Multi-ResNet2026.0487.71
HO-ResNet2026.0487.23
ATONet2026.0486.68
DivBFKDDistillation setting=black-box, Real image budget (N)=2 K, Synthetic image budget (M)=40 K, Resolution=high-resolution2026.0486.48
AdaSTEBackbone=ResNet-18, Mixup=true, Annealing=true2021.1280.91
AdaSTEBackbone=ResNet-18, Mixup=false, Annealing=true2021.1279.66
BayesBiNNBackbone=ResNet-18, Mixup=true2021.1279.59
BayesBiNNBackbone=ResNet-18, Mixup=false2021.1278.19
Standard-KDDistillation setting=black-box, Real image budget (N)=2 K, Synthetic image budget (M)=40 K, Resolution=high-resolution2026.0476.94
Vanilla Adamηinit=10−3, Backbone=ResNet-18, Epochs=20, Seeds=52026.0776.9
Student-AloneReal image budget (N)=2 K, Resolution=high-resolution2026.0476.8
Adam-InitOnly (dir.)ηinit=10−3, Backbone=ResNet-18, Epochs=20, Seeds=5, Probe=Direction-matched, κ=2, Calibration=none2026.0776.7
Adam-InitOnly (dir.)ηinit=10−2, Backbone=ResNet-18, Epochs=20, Seeds=5, Probe=Direction-matched, κ=2, Calibration=none2026.0775
Adam-InitOnly (dir.)ηinit=0.3, Backbone=ResNet-18, Epochs=20, Seeds=5, Probe=Direction-matched, κ=2, Calibration=none2026.0774.2
Adam-InitOnly (dir.)ηinit=10−1, Backbone=ResNet-18, Epochs=20, Seeds=5, Probe=Direction-matched, κ=2, Calibration=none2026.0773.8
Adam-InitOnly (dir.)ηinit=1.0, Backbone=ResNet-18, Epochs=20, Seeds=5, Probe=Direction-matched, κ=2, Calibration=none2026.0772.9
Vanilla Adamηinit=10−2, Backbone=ResNet-18, Epochs=20, Seeds=52026.0772.4
Adam-InitOnly (dir.)ηinit=3.0, Backbone=ResNet-18, Epochs=20, Seeds=5, Probe=Direction-matched, κ=2, Calibration=none2026.0771.8
DIP-KDBackbone=ResNet18, Synthetic Dataset Size=50K2026.0461.84
DFHL-RSBackbone=ResNet18, Synthetic Dataset Size=50K2026.0440.53
IDEALBackbone=ResNet18, Synthetic Dataset Size=50K2026.0436.9
ZSDB3Backbone=ResNet18, Synthetic Dataset Size=50K2026.0433.43
NaiveKDBackbone=ResNet18, Synthetic Dataset Size=50K2026.0433.21