Why does the PSRT_upsample function only use the feature from the second forward propagation, as seen here, while BasicVSR++ uses features from four propagations, as seen here? Was there an experiment conducted that found using only the forward_2 feature to be sufficient or even better than using features from other propagations?
Why does the PSRT_upsample function only use the feature from the second forward propagation, as seen here, while BasicVSR++ uses features from four propagations, as seen here? Was there an experiment conducted that found using only the
forward_2
feature to be sufficient or even better than using features from other propagations?