List your company
ResearchFormerly WizWand

Medical Image Classification on PathMNIST

99.38AccuracyFull dataset
41.056856.198471.3486.4816Feb 20, 2026Mar 7, 2026Mar 23, 2026Apr 8, 2026Apr 23, 2026May 9, 2026May 25, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Full datasetSelection ratio (β)=100%2026.0399.38——
DUCSSelection ratio (β)=30%2026.0398.46——
MedVIT-L2026.0598.498.4—
DUCSSelection ratio (β)=20%2026.0396.7——
MO-MAE2026.0595.399.7—
MedVIT-S2026.0594.299.3—
MedVIT-T2026.0593.899.4—
DUCSSelection ratio (β)=10%2026.0393.49——
BYOLBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0493.36——
SimCLRBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0492.92——
NNCLRBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0492.72——
MoCoV3Backbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0492.5——
Barlow TwinsBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0492.43——
VICRegBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0492.32——
DINOBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0492.03——
ReSSLBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0491.99——
ResNet-50Resolution=282026.0291.199—
EVASelection ratio (β)=30%2026.0391.1——
ResNet-18Resolution=2242026.0290.998.9—
ResNet-182026.0590.998.9—
ResNet-18Resolution=282026.0290.798.3—
TDDSSelection ratio (β)=30%2026.0390.6——
TDDSSelection ratio (β)=20%2026.0390.19——
DUCSSelection ratio (β)=5%2026.0390.02——
ResNet-50Resolution=2242026.0289.298.9—
ResNet-502026.0589.298.9—
Moderate-DSSelection ratio (β)=30%2026.0388.22——
EVASelection ratio (β)=20%2026.0387.53——
Moderate-DSSelection ratio (β)=20%2026.0384.79——
AutoKeras2026.0283.495.9—
AutoKeras2026.0583.495.9—
DUCSSelection ratio (β)=2%2026.0382.91——
EVASelection ratio (β)=10%2026.0382.76——
AMIMV-SSLBackbone=ResNet-50, Initialization=Random, Training Strategy=Self-supervised2026.0482.59——
TDDSSelection ratio (β)=10%2026.0382.51——
BioVLMTrain Params=30K2026.0481.56——
EVASelection ratio (β)=5%2026.0377.9——
TDDSSelection ratio (β)=5%2026.0377.59——
CoOpTrain Params=3K2026.0476.93——
KgCoOpTrain Params=3K2026.0476.61——
PromptSRCTrain Params=69K2026.0476.42——
TCPTrain Params=0.5M2026.0476.21——
CoCoOpTrain Params=44K2026.0475.92——
MaPLeTrain Params=5.3M2026.0475.5——
G-LoG bi-filtration (MLP)Sigma (σ)=02026.0275.395.5—
G-LoG bi-filtration (MLP)Sigma (σ)=0.52026.0275.395.4—
EVASelection ratio (β)=2%2026.0375.19——
Moderate-DSSelection ratio (β)=10%2026.0374.12——
TDDSSelection ratio (β)=2%2026.0373.4——
Google AutoML Vision2026.0272.894.4—
Google AutoML2026.0572.894.4—
BiomedCoOpTrain Params=3K2026.0472.59——
auto-sklearn2026.0271.693.4—
Auto-sklearn2026.0571.693.4—
G-LoG bi-filtration (MLP)Sigma (σ)=12026.0270.894—
G-LoG bi-filtration (MLP)Sigma (σ)=1.52026.0270.293.9—
Topo-Med (MLP)2026.0268.394.2—
Moderate-DSSelection ratio (β)=5%2026.0367.62——
ProGradTrain Params=3K2026.0460.26——
Moderate-DSSelection ratio (β)=2%2026.0352.81——
BioMedCLIPTrain Params=02026.0443.3——
AMIMV-SSLBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—97.96—
Barlow TwinsBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.38—
BYOLBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.44—
DINOBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.28—
EL2NRatio=2%2026.06——36.8
EL2NRatio=5%2026.06——49.9
EVARatio=2%2026.06——53.5
EVARatio=5%2026.06——58.7
FacilityRatio=2%2026.06——77
FacilityRatio=5%2026.06——82.5
Forg.Ratio=2%2026.06——53.5
Forg.Ratio=5%2026.06——58.9
FPSRatio=2%2026.06——73
FPSRatio=5%2026.06——83
Graph Coverage SelectionRatio=2%2026.06——80.9
Graph Coverage SelectionRatio=5%2026.06——85.9
HerdingRatio=2%2026.06——78
HerdingRatio=5%2026.06——84.6
MoCoV3Backbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.27—
NNCLRBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.26—
RandomRatio=2%2026.06——77.5
RandomRatio=5%2026.06——84
ReSSLBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.23—
SimCLRBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.45—
VICRegBackbone=ResNet-50, Initialization=random, Training Protocol=self-supervised2026.04—99.25—