sustechbruce / med-unic Goto Github PK

Hello! I wonder how you got this file: "MIMIC-CXR-meta_INDI_FIND_IMP_report.csv"? I didn't see any code about how to generate this file. Thanks!

Hi, I am trying to reproduce model inference on COVIDx linear probe task, but I can't find the multi-class label file in the COVIDx version 6 dataset. How can I get access to the label file?

I am curious why in the code for downstream fine-tuning tasks (mgca_vit_finetuner.py), when the dataset is Covid, the number of categories is set to 2? Shouldn't it be classified into three categories?

Hi, i am trying to reproduce your code. Could you share your model checkpoint?

Hey, great job!

Could you please provide the relevant code for the attention heat-maps in Figure 5? I'm attempting to visualize numerous pre-training models, but I haven't achieved good results. 😢

Thank you in advance.

I have some confusion about using the dropout strategy for text augmentation, and I did not understand how this was achieved in Med-Unic. Hope to get some guidance.

Hi,

In zero-shot_pdc.py I saw you have 10 disease labels :

cxr_labels: List[str] = ['Atelectasis','Cardiomegaly',
'Consolidation', 'Edema','Pleural Effusion', 'Hilar Enlargement', 'Pneumothorax', 'Pneumonia', 'Vascular Redistribution', 'Heart Insufficiency']

However, in the padchest csv file, there are around 190 unique labels (physician labelled). I am confused about what labels should be used, and hope I can get some clarifications. Cheers!