Siyaqhubeka neendaba zeNkomfa yeTekhnoloji ye-GPU emva kokubhengezwa kwekhompyuter enye UNvidia Jetson Nano ye $ 99 ezinikezelwe ekuphunyezweni kwezicelo kubukrelekrele bokufakelwa kwabaphuhlisi, abaphandi kunye nabazithandayo.
Kwakule GTC 2019 inye, uNvidia, umboneleli wehlabathi weeprosesa kunye neetshipsi zemizobo kutyhilwe umenzi wemifanekiso ophilileyo ngobukrelekrele bokwenziwa Isoftware ibizwa GauGAN ngabaqulunqi bayo, ibonelela ngesishwankathelo samathuba anikezelwa ngamaqonga wenethiwekhi yeNvidia neural.
Le AI yakhela kufundo olusuka kwinkqubo yePix2Pix eyazisiweyo kunyaka ophelileyo Ingabonisa indalo ebonakalayo, utshilo usekela-mongameli kaNvidia wesicelo sokufunda ngokunzulu uBryan Catanzaro, kodwa iPix2Pix ayinakupeyinta umhlaba ngenxa yokuba ukwenza njalo kushiya izinto zobugcisa kumfanekiso obonayo.
I-GauGAN yenzelwe ukwenza umzobo kwaye iwuguqule ube ngumfanekiso we-photorealistic ngemizuzwana. IGauGAN ibonelela ngezixhobo ezithathu: ipheyile, ipeni nepensile.
Umboniso weGauGAN kuhlelo lwangoku lweNkomfa yeTekhnoloji ye-GPU landela ukumiliselwa, phakathi kwinyanga edlulileyo, Ukusuka kwisiza esibonisa imizobo yobuso bomntu eveliswe bubukrelekrele bokuzenzela.
Kufuneka kuqatshelwe ukuba, ekupheleni konyaka ophelileyo, inkampani yayisele ivelise ubukrelekrele bokuzenzela obunakho ukuvelisa ubuso babantu ngento ekhathazayo.
Ingqondo yeGAN
Eyona nto iphambili kula manyathelo ngesoftware yeGauGAN yingqondo yeGAN.
I-GAN yimodeli yokuvelisa apho iinethiwekhi ezimbini zikhuphisana kwimeko yomdlalo wethiyori.
Inethiwekhi yokuqala ngumvelisi, ivelisa isampulu (umzekelo, umfanekiso), Ngelixa umchasi wayo, umkhethi, ezama ukukhangela ukuba isampulu iyinyani na okanye sisiphumo somvelisi.
Ukufunda kunokumodareyitha njengomdlalo we-zero-sum. Ezi nkqubo zekhompyuter zikhuphisana nezigidi zamaxesha ukuphucula ubuchule bakho bokucinga bade babe nakho ukwenza imifanekiso epheleleyo.
Ngamafutshane, i-GAN ithetha ukuba iinethiwekhi ezimbini ziyasebenza ngokuchaseneyo.
Kuqala kufakwa idatha eluhlaza ebolileyo. Ukusuka kwezi, timpuku yenza umfanekiso. Lemva koko yithumele kwenye inethiwekhi ukuba, ineefoto zokwenene okanye imifanekiso kwiziko ledatha yayo. Inethiwekhi yesibini iya kwenza isigwebo somfanekiso kwaye iya kwazisa eyokuqala.
Ukuba umfanekiso awukhangeleki njengeziphumo ezilindelekileyo, ialgorithm yokuqala iyaqhubeka nenkqubo. Ukuba kukho umdlalo, uyaziswa ukuba usendleleni elungileyo kwaye ugqiba ukuqonda ukuba ngowuphi umfanekiso olungileyo.
Ngaba isebenza njalo GauGAN
Nje ukuba uqeqeshwe ngokwaneleyo, unokuvelisa imifanekiso kwityathanga. Ngokwedatha epapashwe nguNvidia, umkhethi osebenza ngasemva kwesoftware yeGauGAN unesiseko sedatha yemifanekiso esisigidi yendalo.
IGauGAN inokubonelela ngesixhobo esinamandla sokwenza indalo ebonakalayo. Nakule demo inemida, kuyacaca ukuba isoftware yakhiwe ngokujikeleza ezi zakhono Ingabhenela wonke umntu ukusuka kubayili bemidlalo yevidiyo ukuya kubakhi bezinto eziqhelekileyo.
Ngobukrelekrele bokuqonda obuqonda ukuba umhlaba wokwenyani ujongeka kanjani, aba baqeqeshi banokubonisa ngcono izimvo zabo kwaye benze utshintsho olukhawulezileyo kwimeko yokwenziwa.
Inkampani ayinazicwangciso zokuyikhupha ithengiselwe urhwebo, kodwa kungekudala inokuqalisa uvavanyo loluntu ukuvumela nabani na ukuba asebenzise isoftware.
Ngokusebenzisa idemo yesoftware yeGauGAN, UNvidia ubalaselisa izinto ezilungileyo zokusetyenziswa yetekhnoloji esekwe kwi-GAN, Kodwa kufuneka kuthiwe olu seto lweetekhnoloji inokusetyenziselwa iinjongo ezimbi.
Deepfakes (imifanekiso eveliswe ngekhompyuter ebekwe ngaphezulu kwezinye okanye iividiyo esele zikho) ziyinxalenye yale ndawo kwaye zithembekile ngabantu besithathu abakhohlakeleyo ukusasaza iindaba ezingeyonyani kunye nokuxoka.
UNvidia ugcina iqonga lokudlala kwi-AI kwi-Intanethi. Uluhlu lweeprojekthi inkampani eyazisa ngazo ngokubhekisele kubukrelekrele bokuzenzela, kwaye abasebenzisi be-intanethi banethuba lokusungula iidemo.
Que lobuhlanga