List your company
ResearchFormerly WizWand

Medical Image Classification on ChestX-ray14

0.8274Mean AUROCREGSL
0.4934560.5801530.666850.753547Nov 8, 2021Jul 23, 2022Apr 7, 2023Dec 20, 2023Sep 3, 2024May 18, 2025Jan 31, 2026
Updated 8mo ago

Evaluation results

MethodLinks
REGSLBackbone=ResNet-182021.110.8274
l2-PGMBackbone=ResNet-182021.110.8235
l2-SPBackbone=ResNet-182021.110.8231
l2-NormBackbone=ResNet-182021.110.8198
SupervisedPretraining Dataset=ImageNet, Backbone=ResNet-50, Evaluation Protocol=Fine-tuning2022.040.817
Fine-tuningBackbone=ResNet-182021.110.8159
DiRAMoCo-v2Pretraining Dataset=ChestX-ray14, Backbone=ResNet-50, Evaluation Protocol=Fine-tuning2022.040.8112
DiRABarlow TwinsPretraining Dataset=ChestX-ray14, Backbone=ResNet-50, Evaluation Protocol=Fine-tuning2022.040.8088
DiRAsimSiamPretraining Dataset=ChestX-ray14, Backbone=ResNet-50, Evaluation Protocol=Fine-tuning2022.040.8044
RandomPretraining Dataset=None, Backbone=ResNet-50, Evaluation Protocol=Fine-tuning2022.040.8031
LSBackbone=ResNet-182021.110.7885
NOVA ViT-BFramework=NOVA, Model=ViT-B, # Parameters=92.1M, # Samples Seen=130K, Zero-Shot=true2026.010.7317
NOVA ViT-SFramework=NOVA, Model=ViT-S, # Parameters=27.1M, # Samples Seen=130K, Zero-Shot=true2026.010.7304
MedCLIP ViT-SFramework=MedCLIP, Model=ViT-S, # Parameters=21.7M, # Samples Seen=130K, Zero-Shot=true2026.010.6795
MedCLIP ViT-BFramework=MedCLIP, Model=ViT-B, # Parameters=86.0M, # Samples Seen=130K, Zero-Shot=true2026.010.667
CLIP (SigLIP) ViT-BFramework=CLIP, Algorithm=SigLIP, Model=ViT-B, # Parameters=150.0M, # Samples Seen=1.41M, Zero-Shot=true2026.010.6639
CLIP (InfoNCE) ViT-BFramework=CLIP, Algorithm=InfoNCE, Model=ViT-B, # Parameters=150.0M, # Samples Seen=1.41M, Zero-Shot=true2026.010.6553
CLIP (Base) ViT-BFramework=CLIP, Algorithm=Base, Model=ViT-B, # Parameters=150.0M, # Samples Seen=1.28M, Zero-Shot=true2026.010.5063