Do you use original label in Stage-2 classifier?

[wjtan99]

Hi, congratulations on your great work in the ICASSP COVID challenge. I read your paper and code. I have some questions regarding your training precdure.

1. In the Stage-2 classifier, in Fig. 3, you draw at the input, normal patient data, predicted 116 covid patient data, and predicted 35 CAP patient data. Did you only use these data to train the Stage-2 classifier? Or do you use these data plus all the labeled slice data for the 55 covid patients and 25 CAP patients?
2. In the Stage-1 classifier you have a separate classifier for COVID and CAP patients. So for patients with patient labels but not slice labels, you use these two classifiers to predict the slice label. And you use these predicted labels in Stage-2. The output of the Stage-2 classifier has three classes. At the input label, do you now mix the COVID-19 non-infection slice label, the CAP non-infection slice label, and the normal patient slice label?
3. In both the Stage-1 and Stage-2 classifiers, you use transfer learning. Have you tried training all parameters including those of the backbone. Did it work badly?

Thanks a lot. Looking forward to your reply.

[wjtan99]

Never mind. I found answers in your README and code.