search

lorenmt / reco

The implementation of "Bootstrapping Semantic Segmentation with Regional Contrast" [ICLR 2022].
https://shikun.io/projects/regional-contrast
Other
162 stars 25 forks source link

Questions about train_u_label #35

Closed sang2232 closed 1 year ago

sang2232 commented 1 year ago

Hello, I am a student studying deep learning. In the code below, I know that train_u_dataset is a data set without label. I wonder what train_u_label is and why it is necessary.

    train_u_data, train_u_label = train_u_dataset.next()
    train_u_data, train_u_label = train_u_data.to(device), train_u_label.to(device)

Thank you for your wonderful code.

lorenmt commented 1 year ago

Hello. These labels are defined in the original dataset, we just don't use them as part of the training. But we can evaluate the quality of generated labels compared to the ground-truth labels. Hope that helps.

sang2232 commented 1 year ago

Thank you for your kind reply.

I understand.