DPAN and train

[Qi-Ran]

Dear Author,

Hello! First of all, thank you very much for your significant contributions. I am very interested in your work and would like to ask you two questions:

What is the significance of the DPAM_layer? I noticed in the paper that it is set to 20, which means it is applied from the 6th layer to the 24th layer. If I change the value to None, does it become the original CLIP image encoder? I also saw your commented note "DPAM_layer = 1, no DPAM is used," what does this mean?

During the training process, saving the weight files for each epoch, and then using the weights from the last epoch directly for testing, how can we ensure that the last epoch's weights are not the best and how to avoid overfitting? Is it necessary to include a validation phase during training to ensure that the best weight file is used for testing?

Best regards.