I-Yandex imemezele ukwethulwa . Igalari izobonisa abasebenzisi imidwebo eyingqayizivele engu-4000, edalwe i-algorithm esekelwe kubuchwepheshe bobuhlakani bokwenziwa. Noma ubani angakhetha omunye wemidwebo esele esitokweni futhi azithathele wona ngokuphelele mahhala. Kulesi simo, umnikazi wayo kuphela ozoba nenguqulo egcwele yomdwebo.
Igalari yobuciko benethiwekhi ye-neural ihlukaniswe ngamakamelo angu-4 anezihloko. Abasebenzisi bangabuka okudaliwe kohlelo lwe-AI ezigabeni zeMvelo, Abantu, Idolobha, kanye Nesimo Sesimo. Igalari ebonakalayo izovumela izivakashi ukuthi zivakashele umbukiso ogcwele ngaphandle kokushiya ikhaya, futhi imisebenzi eziyithandayo zingabiwa nabangane ezinkundleni zokuxhumana.
Izivakashi zokuqala zizokwazi ukulanda isithombe esisodwa ezisithandayo ngosayizi ogcwele. Ukuze ube umnikazi womdwebo owenziwe inethiwekhi ye-neural, uzodinga ukungena kunoma iyiphi isevisi ye-Yandex. Imidwebo etholwa ngabanikazi izoqhubeka nokutholakala ukuze ibukwe, kodwa kugalari izovezwa kuphela ngendlela encishisiwe.
Imisebenzi ethulwe idalwe inethiwekhi ye-neural ephindaphinda ukwakheka kwe-StyleGAN2. Ngenqubo yokuqeqesha inethiwekhi ye-neural, ochwepheshe basebenzisa imisebenzi yezitayela ezahlukene, njenge-cubism, minimalism, ubuciko bomgwaqo, njll. Phakathi nenqubo yokuqeqeshwa, inethiwekhi ye-neural yafunda imidwebo engu-40, ngemva kwalokho yaqala ukudala imisebenzi yayo. Ukuze ukhethe imidwebo ngokwezigaba ezihlukene, kwasetshenziswa enye inethiwekhi ye-neural, esetshenziswa kusevisi ye-Yandex.Izithombe ukucinga izithombe ezisekelwe emibuzweni. Nguye owayekwazi ukubona abantu, imvelo, idolobha kanye nezimo ezihlukahlukene emidwebeni, ehlela imisebenzi etholakalayo ngezigaba.
Source: 3dnews.ru