About HRDA refign #21
Description
hello！
Recently, I have been searching for the state-of-the-art results on the Cityscapes to Darkzurich_night domain adaptation task. So far, the information I have found indicates that the Refign method based on the HRDA architecture still ranks high in the public submissions in terms of test set performance, with a Miou of 63.9. We know that the outstanding contribution of HRDA lies in its context cropping and detail cropping of the source and target domains, achieving a good balance between performance and input resolution. Moreover, your method also includes a reference domain information. I would like to know how you designed the pipeline for such excellent results. More specificly，How did you integrate HRDA into the current Refign pipeline， I would be extremely grateful if you could reply in the form of a diagram.Thanks again！