About labels

[Broccoli77PP]

Thank you for improving your source code, but I would like to ask why the labeled image data is added to the model training in the code, since the labeled image data is not used in the paper

[OMEGA-RS]

Thanks for your question. In fact, only the generated labels from the pretext task are used to train our model in the phase of model building, aiming to obtain the pretrained model to learn useful features of the input data. In the phase of label predict, the label image is used to evaluate the accuracy of the proposed method.

[Broccoli77PP]

Thank you for your prompt response. Can I understand that REF_PATH in My Dataset here does not participate in the final change detection process.

[OMEGA-RS]

Yes, here, the ref_path is only used to load the real label (i.e., the reference image) for evaluating the performance of the proposed method. In the model prediction phase, the real label does not participate.

[Broccoli77PP]

Thank you very much for your reply