Hi @ckkelvinchan, I was wondering how did you choose the weights for the losses of the second training of RealBasicVSR. I mean how did you choose $\lambda{per}$ and $\lambda{adv}$ mentioned in the paper "Investigating Tradeoffs in Real-World Video Super-Resolution"?
$\lambda{per} L{per}$ is almost 2 order of magnitude higher than all other in your log
Hi @ckkelvinchan, I was wondering how did you choose the weights for the losses of the second training of RealBasicVSR. I mean how did you choose $\lambda{per}$ and $\lambda{adv}$ mentioned in the paper "Investigating Tradeoffs in Real-World Video Super-Resolution"? $\lambda{per} L{per}$ is almost 2 order of magnitude higher than all other in your log