Wonke umntu ogcina isilwanyana sasekhaya uyasithanda. Nangona kunjalo, ngaba inja yakho oyithandayo iya kujongeka intle ngakumbi ukuba ibiluhlobo olwahlukileyo? Enkosi kwisixhobo esitsha esivela kwi-NVIDIA esibizwa ngokuba yi-GANimals, unokusivavanya ukuba ingaba isilwanyana osithandayo singajongeka simnandi na ukuba ibisisilwanyana esahlukileyo.
Ekuqaleni kwalo nyaka, uphando lweNVIDIA Abasebenzisi be-Intanethi ngesixhobo sakhe se-GauGAN, esamvumela ukuba ajike imizobo erhabaxa ibe yimifanekiso ephantse ibe yifotorealistic. Esi sixhobo sasifuna abasebenzisi ukuba bachaze ukuba zeziphi iindawo zomfanekiso kufuneka zibe ngamanzi, imithi, iintaba kunye nezinye iimpawu zomhlaba ngokukhetha umbala ofanelekileyo webrashi, kodwa i-GANimals isebenza ngokuzenzekelayo ngokuzenzekelayo. Konke okufuneka ukwenze kukulayisha ifoto yesilwanyana sakho sasekhaya, kwaye kuya kudala uthotho lwemifanekiso yefotorealistic yezinye izilwanyana ezigcina "imbonakalo yobuso" bomzekelo.
Kule veki, kwiphepha elinikezelwe kwiNkomfa yeZizwe ngezizwe kwiKhompyutha yeVision eSeoul, eKorea, abaphandi bachaza i-algorithm abayiphuhlisayo - . Imele ukuGuqulelwa kweMifanekiso-kumfanekiso engagadwanga. Xa usebenzisa ubukrelekrele bokwenziwa ukuguqula iimpawu zomfanekiso womthombo ube ngumfanekiso ekujoliswe kuwo, ubukrelekrele bokwenziwa ngokuqhelekileyo bufuna ukuqeqeshwa kwingqokelela enkulu yemifanekiso ekujoliswe kuyo enamanqanaba okukhanya ahlukeneyo kunye nee-engile zekhamera ukuvelisa iziphumo ezibonakala ziyinyani. Kodwa ukwenza idatabase yomfanekiso omkhulu kangaka kuthatha ixesha elide kwaye kunciphisa amandla enethiwekhi ye-neural. Ukuba i-AI iqeqeshelwe ukuguqula iinkukhu zibe yi-turkeys, yiloo nto kuphela eya kwenza kakuhle.
Ngokuthelekisa, i-algorithm ye-FUNIT inokuqeqeshwa ngokusebenzisa imifanekiso embalwa yesilwanyana esijoliswe kuyo apho siqhutyelwa khona ngokuphindaphindiweyo. Emva kokuba i-algorithm iqeqeshwe ngokwaneleyo, idinga kuphela umfanekiso omnye womthombo kunye nezilwanyana ezijoliswe kuzo, ezinokuthi zibe yinto engabonakaliyo kwaye ayizange iqhutywe okanye ihlalutywe ngaphambili.
Abo banomdla banokuzama iGANanimals kwi , kodwa ukuza kuthi ga ngoku iziphumo ziphantsi kwaye azifanelekanga nantoni na ngaphandle kweenjongo zemfundo okanye ukwanelisa umdla. Abaphandi banethemba lokuphucula amandla e-AI kunye ne-algorithm ukuze kube lula ukutshintsha ubuso babantu ngaphandle kokuxhomekeka kuluhlu olukhulu lwemifanekiso egcinwe ngononophelo.
umthombo: 3dnews.ru