Image Classification on COVIDx
93.5AccuracyMedUnifier (w TIG)
Evaluation results
| Method | Links | |
|---|---|---|
| MedUnifier (w TIG)Training data percentage=100%2025.03 | 93.5 | |
| MedUnifier (w/o TIG)Training data percentage=100%2025.03 | 92.8 | |
| MLIPBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=MIMIC-CXR2024.02 | 92.5 | |
| MGCATraining data percentage=100%2025.03 | 92.3 | |
| MGCABackbone=ViT-B/16, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=MIMIC-CXR2024.02 | 92.3 | |
| ConVIRTTraining data percentage=100%2025.03 | 92 | |
| ConVIRTEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=MIMIC-CXR2024.02 | 92 | |
| MLIPBackbone=ResNet-50, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=MIMIC-CXR2024.02 | 90.8 | |
| MGCABackbone=ResNet-50, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=MIMIC-CXR2024.02 | 90.5 | |
| GLoRIAEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=CheXpert2024.02 | 89 | |
| GLoRIA-MIMICEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=MIMIC-CXR2024.02 | 88.6 | |
| MedUnifier (w TIG)Training data percentage=10%2025.03 | 88.3 | |
| GLoRIATraining data percentage=100%2025.03 | 88 | |
| MedUnifier (w/o TIG)Training data percentage=10%2025.03 | 87.5 | |
| ImageNet InitEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=ImageNet2024.02 | 86.3 | |
| MLIPBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=MIMIC-CXR2024.02 | 86.3 | |
| MLIPBackbone=ResNet-50, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=MIMIC-CXR2024.02 | 85 | |
| MGCATraining data percentage=10%2025.03 | 84.8 | |
| MGCABackbone=ViT-B/16, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=MIMIC-CXR2024.02 | 84.8 | |
| MGCABackbone=ResNet-50, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=MIMIC-CXR2024.02 | 83.5 | |
| ConVIRTTraining data percentage=10%2025.03 | 82.5 | |
| ConVIRTEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=MIMIC-CXR2024.02 | 82.5 | |
| GLoRIA-MIMICEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=MIMIC-CXR2024.02 | 81.5 | |
| GLoRIATraining data percentage=10%2025.03 | 80.5 | |
| ImageNet InitEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=ImageNet2024.02 | 78.8 | |
| GLoRIAEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=CheXpert2024.02 | 77.8 | |
| MedUnifier (w TIG)Training data percentage=1%2025.03 | 76.8 | |
| MedUnifier (w/o TIG)Training data percentage=1%2025.03 | 75.3 | |
| MLIPBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=MIMIC-CXR2024.02 | 75.3 | |
| MGCATraining data percentage=1%2025.03 | 74.8 | |
| MGCABackbone=ViT-B/16, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=MIMIC-CXR2024.02 | 74.8 | |
| MLIPBackbone=ResNet-50, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=MIMIC-CXR2024.02 | 73 | |
| ConVIRTTraining data percentage=1%2025.03 | 72.5 | |
| ConVIRTEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=MIMIC-CXR2024.02 | 72.5 | |
| MGCABackbone=ResNet-50, Evaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=MIMIC-CXR2024.02 | 72 | |
| Random InitEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=100%, Pre-trained Dataset=None2024.02 | 70 | |
| GLoRIAEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=CheXpert2024.02 | 67.3 | |
| GLoRIA-MIMICEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=MIMIC-CXR2024.02 | 67.3 | |
| GLoRIATraining data percentage=1%2025.03 | 66.5 | |
| ImageNet InitEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=ImageNet2024.02 | 64.8 | |
| Random InitEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=10%, Pre-trained Dataset=None2024.02 | 60.3 | |
| Random InitEvaluation Protocol=Fine-tuning, Fine-tuning Data Percentage=1%, Pre-trained Dataset=None2024.02 | 50.5 | |
| MGLLPretrain Data=MIMIC-CXR, Zero-shot evaluation protocol=true2025.11 | 39 | |
| Ark+Pretrain Data=MIMIC-CXR, Zero-shot evaluation protocol=true2025.11 | 37.8 | |
| MGCAPretrain Data=MIMIC-CXR, Zero-shot evaluation protocol=true2025.11 | 37.3 | |
| MAVLPretrain Data=MIMIC-CXR, Zero-shot evaluation protocol=true2025.11 | 37 | |
| FG-CLIPPretrain Data=MIMIC-CXR, Zero-shot evaluation protocol=true2025.11 | 36.3 | |
| RetiZeroPretrain Data=MIMIC-CXR, Zero-shot evaluation protocol=true2025.11 | 35.8 | |
| MedVersaPretrain Data=* (published pretrained model), Zero-shot evaluation protocol=true2025.11 | 35.5 | |
| MLIPBackbone=ViT-B/16, Evaluation Protocol=Zero-shot, Pre-trained Dataset=MIMIC-CXR2024.02 | 34.8 | |
| CheXAgentPretrain Data=* (published pretrained model), Zero-shot evaluation protocol=true2025.11 | 34.3 | |
| MGCABackbone=ViT-B/16, Evaluation Protocol=Zero-shot, Pre-trained Dataset=MIMIC-CXR2024.02 | 33.2 | |
| MLIPBackbone=ResNet-50, Evaluation Protocol=Zero-shot, Pre-trained Dataset=MIMIC-CXR2024.02 | 26.3 | |
| MGCABackbone=ResNet-50, Evaluation Protocol=Zero-shot, Pre-trained Dataset=MIMIC-CXR2024.02 | 24.5 | |
| GLoRIA-MIMICEvaluation Protocol=Zero-shot, Pre-trained Dataset=MIMIC-CXR2024.02 | 22.1 | |
| GLoRIAEvaluation Protocol=Zero-shot, Pre-trained Dataset=CheXpert2024.02 | 20.9 | |
| ConVIRTEvaluation Protocol=Zero-shot, Pre-trained Dataset=MIMIC-CXR2024.02 | 17.8 |