I-VISual AI GUIDE

Ukukhula Okuqhubekayo kwama-GAN

Ukukhula okuqhubekayo kuqeqesha i-GAN ngokuqala ezinqumweni ezincane futhi kancane kancane wengeze izendlalelo ukuze kufinyelelwe ezithombeni ezinokulungiswa okuphezulu.

Uhlolojikelele

Ukukhula okuqhubekayo kuqeqesha i-GAN ngokuqala ezinqumweni ezincane futhi kancane kancane wengeze izendlalelo ukuze kufinyelelwe ezithombeni ezinokulungiswa okuphezulu. Ibalulekile ngoba yenza ukuhlanganiswa kwe-GAN okuzinzile, kwekhwalithi ye-megapixel kwasebenza okokuqala ngqa.

Ukukhula Okuqhubekayo Kwama-GAN okokugeleza komsebenzi okubona ngekhompyutha okuhumusha noma okukhiqiza imidiya ebonakalayo ukuze ihlaziywe, isebenze, nokudala.

I-Deep Dive

Yethulwe nguKarras et al. (i-NVIDIA) ngo-2017, ukukhula okuqhubekayo (i-ProGAN) ibhekana nokungazinzi nokunensa kwama-GAN okuqeqesha ngokuqondile ngokulungiswa okuphezulu. Kokubili ijeneretha kanye nokubandlulula kuqala kuncane, ngamaphikseli angu-4x4, ukufunda isakhiwo esikhulu kuphela. Izendlalelo ezintsha eziphinda kabili ukulungiswa (8x8, 16x16, kufika ku-1024x1024) bese zengezwa ngokulinganayo kuwo womabili amanethiwekhi ngesikhathi sokuqeqeshwa. Okubalulekile ukuthi, ungqimba ngalunye olusha lufiphala ngokushelela ekusebenziseni ukuhlanganisa kwe-alpha komugqa ukuze inethiwekhi ingashaqiswa ukuguqulwa kwezakhiwo okungazelelwe. Ngokufunda izici ezimahhadla ngaphambi kwemininingwane emihle, ukuqeqeshwa kuzinze kakhudlwana, kuhlangana ngokushesha, futhi kukhiqiza ubuso obuthembekile obenze imiphumela ye-CelebA-HQ yaduma. Leli phepha liphinde lethula ukuchezuka okujwayelekile kwe-minibatch kanye namazinga okufunda alinganayo ukuze kuqhutshekwe kuzinziswe ukuqeqeshwa.

I-Technical Insight

I-fade-in iyisu eliphakathi. Uma ibhulokhi enokulungiswa okuphezulu kwengezwa, okukhiphayo kuhlanganiswa nenguqulo yesampula yokulungiswa kwangaphambilini kusetshenziswa i-alpha yesisindo ephuma ku-0 iye ku-1. Lokhu kuvumela izisindo zezendlalelo ezintsha ukuthi zifudumale kancane kancane esikhundleni sokuphazamisa lokho inethiwekhi esekufundile. Inqubo ye-symmetric yenzeka kumbandlululi. Ukuchezuka okujwayelekile kwe-Minibatch kwenezela isici esifingqa ukuhluka kwenqwaba, kungakhuthazi ijeneretha ukuthi ingagoqeki kuya kokuphumayo okulinganiselwe.

Ukuphumelela Ukukhula Okuqhubekayo kwama-GAN

Ukuze wakhe ukuqonda okujulile, phatha i-Progressive Growing of GANs njengemodeli yokusebenza, hhayi isici esisodwa. Chaza imiphumela oyifunayo, cacisa ukucabanga, futhi uhlukanise lokho isistimu engakwenza ngokwethembeka kulokho okusadinga ukwahlulela kochwepheshe.

Empeleni, amaqembu aqinile asebenzisa ukunemba kwebhalansi ye-Progressive Growing of GANs namaqiniso okusebenza njengekhwalithi yedatha, ukuhluka kokukhanya, nokuvumelana kwamalebula. Babhala imibandela yempumelelo ecacile, ukuhlola okuqhathaniswa nedatha engokoqobo nokugeleza komsebenzi, futhi baphindaphinde ngokusekelwe kumaphethini okuhluleka aqashiwe esikhundleni sokuwina kwebhentshimakhi yesikhathi esisodwa. Yilapho ukuqonda kwethiyori kuguquka kube amandla ahlala njalo kuwo wonke umkhiqizo, inqubomgomo, kanye nokusebenza.

I-Visual AI ingakwazi ukuhlola, ukutholwa, nokumaka imisebenzi esikalini. Ngesikhathi esifanayo, amalungelo ezithombe kanye nemvume kungaba ubungozi bomthetho uma ukutholakala kungacacile. Indlela eqine kakhulu iwukuhlanganisa isivinini sokuhlola nesiyalo sokuphatha: qhuba abashayeli bezindiza, bamba ubufakazi, ushicilele amalogi ezinqumo, futhi ubuyekeze izivikelo ngokuqhubekayo njengoba imodeli yokuziphatha, okulindelwe ngabasebenzisi, kanye nezimfuneko zokulawula zishintsha.

I-Strategic Impact

I-Visual AI ingakwazi ukuhlola, ukutholwa, nokumaka imisebenzi esikalini.

I-Visual AI ingakwazi ukuhlola, ukutholwa, nokumaka imisebenzi esikalini. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.

Amathimba aqanjiwe angakwazi ukulinganisa imiqondo ngokushesha ngezibuyekezo ezimbalwa ezenziwa mathupha.

Amathimba aqanjiwe angakwazi ukulinganisa imiqondo ngokushesha ngezibuyekezo ezimbalwa ezenziwa mathupha. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.

Imisebenzi ingasebenzisa amasiginali wesithombe nawevidiyo obekunzima ukuwenza ngaphambilini.

Imisebenzi ingasebenzisa amasiginali wesithombe nawevidiyo obekunzima ukuwenza ngaphambilini. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.

Ikusasa Lokukhula Okuqhubekayo kwama-GAN

Ukukhula okuqhubekayo kwaba isisekelo se-StyleGAN esakhelwe phezu kwaso, kodwa i-StyleGAN2 kamuva yabonisa ukuthi i-architecture engaguquki enoxhumo olweqayo namabhulokhi ayinsalela ingase ifane nekhwalithi yayo ngaphandle kweshejuli ehleliwe, ngakho ukukhula okusobala akuzange kuthandeke. Ifa elijulile liyaphikelela: isizukulwane se-coarse-to-fine manje sesivela ekusakazeni ngezikali eziningi, amapayipi okucaca okungaphezulu kwe-cascade, kanye nezinsimbi ezikhuphukayo zesikhala. Ukuqonda ukukhula okuqhubekayo kuhlala kubalulekile ekuqondeni ukuthi kungani ukufunda kwe-hierarchical, low-to-high-frequency kuzinza ukuqeqeshwa kokukhiqiza.

Ukuqaliswa Komhlaba Wangempela

Ikhiqiza izithombe zobuso be-CelebA-HQ ezinokulungiswa okuphezulu ezibonisa ukuhlanganiswa kwe-GAN okungu-1024x1024.

Ikhiqiza amasampula ekhwalithi ephezulu kwezinye izizinda njengamagumbi okulala (LSUN) nezinto ngesilinganiso.

Isebenza njengendawo yokuqala yezakhiwo eyanwetshwa i-StyleGAN ukuze ikhiqize ubuso obulawulekayo.

Ukufundisa umgomo wokuqeqesha ukusuka kwe-coarse-to-fine ophinde wasetshenziswa kumapayipi akhiqizayo anezilinganiso eziningi.

Amaphethini Okusebenzisa

Ukukhula Okuqhubekayo kwama-GAN ekusebenzeni

Ikhiqiza izithombe zobuso be-CelebA-HQ ezinokulungiswa okuphezulu ezibonisa ukuhlanganiswa kwe-GAN okungu-1024x1024.

Amaqembu ngokuvamile athola imiphumela engcono uma echaza izilinganiso zekhwalithi ngaphambili, agcina indlela yokukhuphuka yomuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

Ukukhula Okuqhubekayo kwama-GAN ekusebenzeni

Ikhiqiza amasampula ekhwalithi ephezulu kwezinye izizinda njengamagumbi okulala (LSUN) nezinto ngesilinganiso.

Amaqembu ngokuvamile athola imiphumela engcono uma echaza izilinganiso zekhwalithi ngaphambili, agcina indlela yokukhuphuka yomuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

Ukukhula Okuqhubekayo kwama-GAN ekusebenzeni

Isebenza njengendawo yokuqala yezakhiwo eyanwetshwa i-StyleGAN ukuze ikhiqize ubuso obulawulekayo.

Amaqembu ngokuvamile athola imiphumela engcono uma echaza izilinganiso zekhwalithi ngaphambili, agcina indlela yokukhuphuka yomuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

Ukukhula Okuqhubekayo kwama-GAN ekusebenzeni

Ukufundisa umgomo wokuqeqesha ukusuka kwe-coarse-to-fine ophinde wasetshenziswa kumapayipi akhiqizayo anezilinganiso eziningi.

Amaqembu ngokuvamile athola imiphumela engcono uma echaza izilinganiso zekhwalithi ngaphambili, agcina indlela yokukhuphuka yomuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.

Izingozi & Guardrails

!

Amalungelo ezithombe kanye nemvume kungaba ubungozi bezomthetho uma ukuvela kungacacile.

!

Ukusebenza kwemodeli kungahluka kukho konke ukukhanya, izibalo zabantu, kanye nezindawo.

!

Okuhle okungelona iqiniso kungase kungabonakali ngaphandle uma izinga lokuzethemba liqashelwa.

Ukuqalisa Umhlahlandlela

1

Chaza indlela yokwamukela yokunemba, ukukhumbula, nezindleko zamaphutha.

Phatha lokhu njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

2

Hlola ngedatha efana nezimo zangempela zokukhiqiza.

Phatha lokhu njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

3

Engeza isibuyekezo somuntu ukuze uthole ukuzethemba okuphansi noma izibikezelo zomthelela omkhulu.

Phatha lokhu njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

4

Landelela ukukhukhuleka kwemodeli bese uqinisekisa kabusha ngemva kwezinguquko zekhamera noma zesethi yedatha.

Phatha lokhu njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.

Qhubeka Uhlole

I-Computer Vision

Qonda amasistimu ayisisekelo anika amandla i-AI ebonakalayo.

Funda Umhlahlandlela

I-AI Image Generation

Hlola ukugeleza komsebenzi wokudala kanye nokuhwebelana kwamamodeli.

Funda Umhlahlandlela

Check your understanding

Test yourself: take the Progressive Growing of GANs quiz

Start quiz →

Related guides