Some questions about the implementation details

[littleYaang]

Thanks for your excellent work ARIS. There is some questions about the implementation details of the results in paper. In the Table 1, ARIS choose the IPT, SwinIR, HAT as the baseline network and use their pre-trained model on scale x2. "Each resulting model is trained individually." I am curious that for scale 2, why is the performance not as good as the original model after using the plugin ? The analysis of "Compare to Baseline SR Network" says that the baseline SR networks are trained for a specifc scale ,but our ARIS may use the same implementation details as these baseline, that is for a specific network such as SwinIR-ARIS, the results of x2, x3, x4, x6, x8 is trained sperately. Meanwhile, looking forward to your release of the code.