Some Quesions

[hhhhhhalf]

Hello, thank you for your excellent work! I would like to ask about the hierarchical approach used in your paper to expand the window size. From the code, it seems that the shifted window operation was not employed. Does the interaction of information between windows rely on the combination of channel self-correlation and spatial self-correlation in "SCC" to compensate for this? Do you think it would be appropriate to consider incorporating the shifted window operation?

[XiangZ-0]

Hi, thanks for your interest! Yes, there is no window shifting operation in our method, and HiT-SR mainly relies on multi-scale large windows to directly expand the receptive field (the comparisons of receptive fields can be found in Fig. 7 of our paper). I think bringing the window shifting back to HiT-SR might contribute to the performance, but the inference speed will be slower (please see the rightmost image of Fig. 1b for the speed comparison, where the blue bars used window shifting operations).
Hope this helps!

[hhhhhhalf]

Okay, I understand now. Thank you for your reply!