Overview
StyleGAN inobereka mhandu network kubva kuNVIDIA inogadzira zviso zvinokatyamadza zviso uye zvinhu nekubaya majekiseni eruzivo pane yega yega. Izvo zvine basa nekuti dhizaini yaro inopa zvisati zvamboitika, zvakapatsanurwa kutonga pamusoro pehukasha uye yakanaka mufananidzo hunhu.
StyleGAN Architecture ndeyekombuta-chiratidzo workflows inodudzira kana kuburitsa midhiya yekuona yekuongorora, mashandiro, uye kugadzira.
Deep Dive
StyleGAN, yakaunzwa naKarras et al. mu 2018, yakagadzirisazve GAN jenareta pamusoro pepfungwa ye 'style.' Panzvimbo yekudyisa vhekitari isina kurongeka yakananga kunetiweki, inotanga kumepu yakavanzika kodhi z kuburikidza ne8-layer MLP munzvimbo yepakati W, iyo inoparadzanisa zvinhu zvekusiyana. Iyo yakadzidziswa inogara ichitemerwa inozosimudzwa zvishoma nezvishoma, uye pakugadziriswa kwega kwega vheti yemaitiro inogadzirisa mamepu kuburikidza neAdaptive Instance Normalization (AdaIN), inodzora hunhu kubva kupose (yakakora maseru) kuenda kune ganda mavara (akanaka akaturikidzana). Per-layer ruzha zvinopinza zvinowedzera stochastic zvakadzama senge mafinya uye bvudzi rakarasika. StyleGAN2 (2020) yakatsiva AdaIN nekuremerwa kwehuremu kuti ibvise 'blob' zvigadzirwa, uye StyleGAN3 (2021) yakagadziriswa mameseji-inonamira aliasing kuita kuti maficha afambe zvakangoitika panguva yeanimation.
Technical Insight
Iyo yakakosha michina ndeye-style-based modulation. Iyo network yemepu inoshandura z kuita w, uye yakadzidzwa affine inoshandura w kuita chikero che-per-channel uye kurerekera kunoshandiswa kune akajairwa mamepu ezvimiro pachisarudzo chega chega. Nekuti masitayera anoita layer-ne-layer, unogona kusanganisa w yemufananidzo payakakombama neimwe pamitsetse yakanaka ('style kusanganisa') kuchinjana chimiro uchichengeta magadzirirwo. StyleGAN2's demodulation inopeta idzi manhamba muhuremu hwe convolution, ichibvisa normalization artifacts.
Mastering StyleGAN Architecture
StyleGAN inobereka mhandu network kubva kuNVIDIA inogadzira zviso zvinokatyamadza zviso uye zvinhu nekubaya majekiseni eruzivo pane yega yega. Izvo zvine basa nekuti dhizaini yaro inopa zvisati zvamboitika, zvakapatsanurwa kutonga pamusoro pehukasha uye yakanaka mufananidzo hunhu. StyleGAN Architecture ndeyekombuta-chiratidzo workflows inodudzira kana kuburitsa midhiya yekuona yekuongorora, mashandiro, uye kugadzira. Kuvaka kunzwisisa kwakadzama, bata StyleGAN Architecture semuenzaniso wekushandisa, kwete chinhu chimwe chete: tsanangura zvaunoda, tsanangura fungidziro, uye patsanura zvinogona kuitwa nehurongwa hwakavimbika kubva kune zvichiri kuda kutonga kwenyanzvi.
Mukuita, zvikwata zvakasimba zvinoshandisa StyleGAN Architecture chiyero chechokwadi nemashandiro anoita semhando yedata, kusiyana kwemwenje, uye kuenderana kwemazita. Ivo vanonyora zvakajeka maitiro ebudiriro, bvunzo vachipokana ne data rechokwadi uye mafambiro ebasa, uye iterate zvichibva pane zvakacherechedzwa maitiro ekutadza kwete kuhwina-nguva imwe chete yebhenji. Apa ndipo apo kunzwisisa kwe theoretical kunoshanduka kuve kugona kwakasimba pane chigadzirwa, mutemo, uye mashandiro.
Visual AI inogona kuita otomatiki yekuongorora, yekuona, uye yekumaka mabasa pachiyero. Panguva imwecheteyo, kodzero dzeMufananidzo uye kubvumirwa kunogona kuve njodzi dzepamutemo kana hunhu husina kujeka. Nzira yakatsiga ndeyekubatanidza kukurumidza kuyedza nekutonga: mhanyisa vatyairi vendege, tora humbowo, buritsa matanda esarudzo, uye urambe uchivandudza chengetedzo semaitiro emuenzaniso, zvinotarisirwa nemushandisi, uye zvinodikanwa zvekutonga.
Strategic Impact
Visual AI inogona kuita otomatiki yekuongorora, yekuona, uye yekumaka mabasa pachiyero.
Visual AI inogona kuita otomatiki yekuongorora, yekuona, uye yekumaka mabasa pachiyero. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Zvikwata zvekugadzira zvinogona prototype pfungwa nekukurumidza nekudzokororwa kwemaoko mashoma.
Zvikwata zvekugadzira zvinogona prototype pfungwa nekukurumidza nekudzokororwa kwemaoko mashoma. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Mashandisirwo anogona kushandisa masaini emifananidzo nemavhidhiyo ayo aimbove akaoma kugadzirisa.
Mashandisirwo anogona kushandisa masaini emifananidzo nemavhidhiyo ayo aimbove akaoma kugadzirisa. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Real-World Implementation
Kugadzira zvisingaperi zviso zvemunhu, zviso zvisipo, sezvakaratidzwa ne thispersondoesnotexist.com.
Semantic face editing: kunyatsochinja zera, kutaura, kana kumira nekufamba nenzira munzvimbo yeW.
Kugadzira dhizaini yekudzidzira data uye maavatars kana chaiwo, zvakavanzika-zvakachengeteka mifananidzo iri kushomeka.
Maturusi ehunyanzvi anopindirana kana 'maitiro-musanganiswa' pakati pemifananidzo kusanganisa chimiro chakakasharara uye ruzivo rwakanaka.
Maitiro Ekuita
StyleGAN Architecture mukuita
Kugadzira zvisingaperi zviso zvemunhu, zviso zvisipo, sezvakaratidzwa ne thispersondoesnotexist.com.
Kugadzira kusingaperi mafotorealistic, zviso zvisipo zvevanhu, sekuratidzwa kwazvinoitwa thispersondoesnotexist.com Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi emupendero, uye kuteedzera zvese zvakawanikwa zvechigadzirwa uye mutengo wekukanganisa nekufamba kwenguva.
StyleGAN Architecture mukuita
Semantic face editing: kunyatsochinja zera, kutaura, kana kumira nekufamba nenzira munzvimbo yeW.
Semantic face editing: kunyatsochinja zera, kutaura, kana kumira nekufamba nenzira muW space Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
StyleGAN Architecture mukuita
Kugadzira dhizaini yekudzidzira data uye maavatars kana chaiwo, zvakavanzika-zvakachengeteka mifananidzo iri kushomeka.
Kugadzira yekugadzira data yekudzidzira uye maavatars kana chaiwo, kuvanzika-akachengeteka mifananidzo iri kushomeka Zvikwata zvinowanzowana mhedzisiro iri nani kana vachitsanangudza zvemhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa uye mutengo wekukanganisa nekufamba kwenguva.
StyleGAN Architecture mukuita
Maturusi ehunyanzvi anopindirana kana 'maitiro-musanganiswa' pakati pemifananidzo kusanganisa chimiro chakakasharara uye ruzivo rwakanaka.
Zvishandiso zveunyanzvi zvinodudzira kana 'maitiro-musanganiswa' pakati pemifananidzo kuti isanganise chimiro chakaomarara uye neruzivo rwakanaka Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura hunhu hwepamberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
Njodzi & Guardrails
Kodzero dzemifananidzo uye kubvumirwa kunogona kuve njodzi dzepamutemo kana provenance isina kujeka.
Kuita kwemuenzaniso kunogona kusiyanisa kupenya, huwandu hwevanhu, uye nharaunda.
Manyepo enhema anogona kusacherechedzwa kunze kwekunge zvikumbaridzo zvekuvimba zvikatariswa.
Implementation Roadmap
Tsanangura maitiro ekugamuchirwa echokwadi, kurangarira, uye mutengo wekukanganisa.
Tsanangura maitiro ekugamuchirwa echokwadi, kurangarira, uye mutengo wekukanganisa. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Edzai nedata rinoenderana nemamiriro chaiwo ekugadzira.
Edzai nedata rinoenderana nemamiriro chaiwo ekugadzira. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Wedzera ongororo yemunhu kune yakaderera-kusavimbika kana yakakwirira-inokanganisa kufanotaura.
Wedzera ongororo yemunhu kune yakaderera-kusavimbika kana yakakwirira-inokanganisa kufanotaura. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Tevera modhi kudonha uye simbisa mushure mekuchinja kwekamera kana dataset.
Tevera modhi kudonha uye simbisa mushure mekuchinja kwekamera kana dataset. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.