ad 5) To je skoro spravne, ale neprovadi se tam zadne dopocitavani. DNN maji takzvany input layer a output layer a mezi nimi jsou hidden layers.
DLSS bude mit rekneme 2k input a 4k ouput layer, DLSS 2x bude mit 4k input a 4k output layer. Vysledna data se ctou uz primo z output layeru.
" In this case, DLSS input is rendered at the final target resolution and then combined by a larger DLSS network".
Neboli presne jak rika Huang v keynote, co jsem postnul, ta DNN se nauci pochopit, co je ve scene a uz primo vi, jaky pixel ma jit na dane misto.
„It can then generate pixels in has never seen before, pixels that make sense to go there."