I really appreciate you providing the code. It has been very helpful for my research. @jiankang1991

I was wondering if it would be possible to get the weight that produce the scores mentioned in the paper.
I have tried several times using the provided code, but I have not been able to reach the scores mentioned in the paper.
If it is not a problem, could I receive the weight?

Thank you.

@jiankang1991 Thank you for providing the script.
can you please provide test example script if possible?

I have read your paper and would like to respectfully inquire whether my understanding of it is correct.
I have some questions that I would like to ask for clarification about Distillation Loss

I hope you don't mind me asking, but from my understanding of the formula and code presented in the paper, it seems that only the SAR-Optical Shared Encoder is trained using the distillation loss during the second training. Additionally, it appears that the distillation loss is calculated solely based on the results obtained from the Optical Stage(pre-trained Optical network and SAR train network) and not from the SAR Stage. Could you please confirm if my understanding is correct?

If my understanding is correct, would this mean that the SAR Stage is not affected by the Distillation Loss during backpropagation?
(And I assume you did this because the features of SAR and Optical networks become increasingly dissimilar at higher levels.)

I have attached the code for your reference as well in your code.
dist_rgb_loss = sum([self.dist_f_loss(sar_f, rgb_f) for sar_f, rgb_f in zip(sar_features, rgb_features)])
sar_features : Optical Stage features from second Training Process (SAR+RGB training process)
rgb_featuers : Optical Stage features from pre-trained Optical Network

Thanks