Siyaqhubeka nezindaba zeNgqungquthela ye-GPU Technology ngemuva kokumenyezelwa kwekhompyutha eyodwa UNvidia Jetson Nano of $ 99 anikezelwe ukuqaliswa kwezicelo ku-intelligence yokufakelwa yabathuthukisi, abacwaningi kanye nama-hobbyists.
Kule GTC 2019 efanayo, uNvidia, umhlinzeki womhlaba wonke wamaprosesa nama-chip graphics wembula umenzi wesithombe ophakanyiswa ubuhlakani bokufakelwa. Isoftware ibizwa I-GauGAN ngabaqambi bayo, ihlinzeka ngokubuka konke ngamathuba ahlinzekwa ngamapulatifomu wenethiwekhi yeNvidia.
Le AI yakhela kokufundwayo okuvela ohlelweni lwePix2Pix olwethulwe ngonyaka odlule lokho kungamela imihlaba ebonakalayo, kusho iphini likamongameli kaNvidia ocwaningweni lokufunda olujulile uBryan Catanzaro, kepha iPix2Pix ayikwazi ukudweba izindawo ngoba ukwenza kanjalo kushiya izinto zobuciko esithombeni.
I-GauGAN yenzelwe ukwenza umdwebo bese iwuguqula ube isithombe se-photorealistic ngemizuzwana. IGauGAN inikela ngamathuluzi amathathu: ibhakede lokupenda, ipeni nepensela.
Umbukiso waseGauGAN kumagazini wamanje weNgqungquthela ye-GPU Technology landela ukwethulwa, maphakathi nenyanga edlule, kusuka kusayithi ekhombisa izithombe zobuso babantu ezakhiwe ubuhlakani bokufakelwa.
Kumele kuqashelwe ukuthi, ekupheleni konyaka owedlule, inkampani yayivele iveze ubuhlakani bokufakelwa obukwazi ukudala ubuso babantu ngeqiniso elikhathazayo.
Umqondo we-GAN
Isici esivamile salezi zinhlelo nge-software yeGauGAN ngumqondo we-GAN.
I-GAN iyimodeli yokuzala lapho amanethiwekhi amabili ancintisana esimweni sethiyori yomdlalo.
Inethiwekhi yokuqala ingu-generator, kukhiqiza isampula (ngokwesibonelo, isithombe), ngenkathi imbangi yayo, umbandlululi, izama ukuthola ukuthi isampula ingeyangempela yini noma uma ingumphumela we-generator.
Ukufunda kungenziwa njengemidlalo eyi-zero-sum. Lezi zinhlelo zamakhompiyutha zincintisana nezigidi zezikhathi ukuthuthukisa amakhono akho okucabanga baze babe namandla okwenza izithombe eziphelele.
Kalula nje, i-GAN isho ukuthi amanethiwekhi amabili ayasebenzisana.
Kuqala kufakwa idatha eluhlaza ebolile. Kusukela kulokhu, tigundane lakha isithombe. Lbese uyithumela kwenye inethiwekhi lokho, inezithombe zangempela noma izithombe ku-database yayo. Le nethiwekhi yesibili izokwenza ukwahlulela kwesithombe futhi izokwazisa eyokuqala.
Uma isithombe singabonakali njengomphumela olindelekile, i-algorithm yokuqala iqala kabusha inqubo. Uma kunomdlalo, waziswa ukuthi usendleleni efanele futhi ugcina ngokuqonda ukuthi isithombe esihle siyini.
Ngabe isebenza kanjani I-GauGAN
Uma usuqeqeshwe ngokwanele, ungakhiqiza izithombe kuketanga. Ngokuya ngemininingwane eshicilelwe nguNvidia, ukubandlulula okusebenza ngemuva kwesoftware yeGauGAN kunesizinda semininingwane yesigidi semvelo.
IGauGAN ingahlinzeka ngethuluzi elinamandla lokwakha imihlaba ebonakalayo. Ngisho nakule demo elinganiselwe, kuyacaca ukuthi isoftware yakhelwe eduze kwala makhono kungakhanga kuwo wonke umuntu kusuka kubakhi begeyimu yevidiyo kuya kubakhi bezakhiwo kuya kubadlali bamageyimu abangajwayelekile.
Ngobuhlakani bokuzenzela obuqonda ukuthi umhlaba wangempela ubukeka kanjani, labo ngoti bangaveza kangcono imibono yabo futhi benze izinguquko ezisheshayo endaweni yokwenziwa.
Le nkampani ayinazo izinhlelo zokuyikhipha ezentengiselwano, kepha kungekudala izokwethula isivivinyo somphakathi sokuvumela noma ngubani ukuthi asebenzise isoftware.
Ngokusebenzisa idemo yesoftware yeGauGAN, UNvidia uqokomisa okuhle kokusetshenziswa kobuchwepheshe obususelwa ku-GAN, Kepha kufanele kuthiwe leli sethi lamasu lingasetshenziselwa izinhloso ezimbi.
Deepfakes (izithombe ezenziwe ngamakhompiyutha zibekwe kwamanye noma kumavidiyo akhona) ziyingxenye yalesi sici futhi zethenjwa ngabantu besithathu abanonya ukusabalalisa izindaba ezingezona nezamanga.
UNvidia ugcina i-AI Playground online platform. Kubala amaphrojekthi inkampani eyethula kuwo ngokobuhlakani bokuzenzela, futhi abasebenzisi be-inthanethi banethuba lokwethula amademo.
Kuhle kanjani