StyleGAN Architecture
StyleGAN bụ netwọk mmegide sitere na NVIDIA na-emepụta ihu na ihe dị egwu site na itinye ozi ụdị n'ụdị ọ bụla.
Nchịkọta
It matters because its design gives unprecedented, disentangled control over coarse and fine image attributes.
Ime miri emi
StyleGAN, nke Karras et al webatara. na 2018, megharịrị GAN generator gburugburu echiche nke 'style'. Kama ịzụ vector na-enweghị usoro ozugbo na netwọkụ, ọ na-ebu ụzọ na-esetịpụ koodu latent z site na MLP 8-layer n'ime oghere etiti W, nke na-ewepụ ihe dị iche iche. A na-eji nke nta nke nta na-etulite tensor amụtara mgbe niile, na mkpebi ọ bụla ụdị vector na-emezigharị maapụ atụmatụ site na Nhazi Nhazi (AdaIN), na-achịkwa njirimara sitere na pose (n'ígwé ndị siri ike) na udiri anụ ahụ (ezigbo n'ígwé). Ntinye mkpọtụ nke otu oyi akwa na-agbakwunye nkọwa stochastic dị ka freckles na ntutu isi kpafuo. StyleGAN2 (2020) dochie AdaIN na ịdị arọ dị arọ iji wepụ ihe 'blob' artifacts, yana StyleGAN3 (2021) edoziri ederede ederede iji mee ka njirimara na-aga n'ụzọ nkịtị n'oge ihe nkiri.
Nghọta nka nka
Isi usoro bụ modulation dabere na ụdị. Netwọk nke eserese ahụ na-atụgharị z ka ọ bụrụ w, na affine mụtara na-agbanwe agbanwe w ka ọ bụrụ ọnụ ọgụgụ ọwa ọ bụla yana nhụsianya etinyere na maapụ atụmatụ ahaziri ahazi na mkpebi ọ bụla. N'ihi na ụdị na-eme Layer-by-layer, ị nwere ike ịgwakọta w nke otu ihe oyiyi na akwa akwa na nke ọzọ na mma n'ígwé ('ịgwakọta ụdị') iji gbanwee pose mgbe ị na-edobe udidi. Ntugharị nke StyleGAN2 na-atụgharị ọnụ ọgụgụ ndị a n'ime oke convolution, na-ewepụ ihe arụrụ arụ ọrụ.
Mmetụta atụmatụ
Ọsọ na ọnụ ọgụgụ
Visual AI nwere ike megharịa nyocha, nchọpụta na mkpado ọrụ n'ọtụtụ.
Mee nhọrọ
Otu ndị na-emepụta ihe nwere ike imepụta echiche ngwa ngwa site na ngbanwe akwụkwọ ntuziaka ole na ole.
Team na usoro ọrụ
Ọrụ nwere ike iji onyonyo na akara vidiyo siri ike ịhazi.
Ọdịnihu nke StyleGAN Architecture
Ọ bụ ezie na ụdị mgbasa ozi ugbu a na-eduga n'ọgbọ ederede gaa n'ọgbọ, StyleGAN nwere ahaziri nke ukwuu, oghere latent edezi (W na W +) na-eme ka ọ bụrụ etiti n'ihu edezi ihu, njikwa njirimara, yana njikọ oge ebe GAN na-adị ngwa ngwa. Na-atụ anya ka ọrụ na-aga n'ihu na ntụgharị GAN (na-emepụta ezigbo foto n'ime W), ụdị dị iche iche nke 3D dị ka EG3D na-enye echiche na-agbanwe agbanwe, yana ngwakọ na-ejikọta latent nke StyleGAN na-achịkwa ya na mgbasa ozi ma ọ bụ mgbanwe mgbanwe maka ụwa abụọ kachasị mma.
Mmejuputa n'ezie n'ụwa
Na-emepụta ihu mmadụ na-adịghị agwụ agwụ, nke na-adịghị adị, dị ka e gosipụtara site na thispersondoesnotexist.com.
Semantic ihu edezi: were na-agbanwe afọ, okwu, ma ọ bụ pose site na-aga n'ihu na ntụzịaka na W oghere.
Ịmepụta data ọzụzụ sịntetik na avatars mgbe ọ dị adị, onyonyo nchekwa nzuzo dị ụkọ.
Ngwá ọrụ nka nke na-ejikọta ma ọ bụ 'ịgwakọta' n'etiti onyonyo iji gwakọta nhazi na ezigbo nkọwa.
Ihe ize ndụ & okporo ụzọ nche
Ikike onyonyo na nkwenye nwere ike bụrụ ihe egwu dị n'iwu ma ọ bụrụ na edoghị anya.
Ọrụ nlereanya nwere ike ịdịgasị iche n'ofe ọkụ, igwe mmadụ, na gburugburu.
Enwere ike ghara ịhụ ihe dị mma ma ọ bụrụ na enyochaghị oke ntụkwasị obi.
Map mmejuputa
Kọwaa ụkpụrụ nnabata maka nkenke, icheta, na ụgwọ njehie.
Nwalee na data dabara na ọnọdụ mmepụta n'ezie.
Tinye nyocha mmadụ maka obere obi ike ma ọ bụ amụma mmetụta dị elu.
Sochie ihe nlere anya wee megharịa ka emechara mgbanwe igwefoto ma ọ bụ dataset.
Nọgide na-eme nchọpụta
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the StyleGAN Architecture quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ntuziaka na-esote
U-Net Architecture
Ajụjụ a na-ajụkarị
What is StyleGAN Architecture?
StyleGAN bụ netwọk mmegide sitere na NVIDIA na-emepụta ihu na ihe dị egwu site na itinye ozi ụdị n'ụdị ọ bụla. Ọ dị mkpa n'ihi na imewe ya na-enye njikwa a na-enwetụbeghị ụdị ya, nke a na-ekewaghị ekewa n'ime àgwà oyiyi mara mma na nke mara mma.
Kedu ebumnuche nke netwọk maapụ StyleGAN?
Maapụ MLP 8-layer z ruo W, oghere dị n'etiti ebe ihe dị iche iche na-ekewa nke ọma, na-enye njikwa dị ọcha.
Na StyleGAN izizi, kedu ka esi etinye ụdị n'ime maapụ atụmatụ?
AdaIN na-ahazi maapụ atụmatụ na-ahazi wee tụọ nha ma tụgharịa ya site na iji ọnụ ọgụgụ sitere na ụdị na mkpebi ọ bụla.
Kedu ihe netwọk njikọ na-amalite kama ịbụ vector latent?
StyleGAN na-amalite site na ntinye a na-amụta mgbe niile ma na-agbanye ụdị na mkpọtụ ka ọ na-agbago, kama ịzụ z na ozugbo.
Kedu ihe ngbanwe ụdị n'ígwé ndị siri ike (obere mkpebi) bụ isi na-achịkwa?
N'ígwé ndị na-adịghị ahụkebe na-achịkwa ihe owuwu buru ibu dị ka pose na ọdịdị n'ozuzu ya, ebe ọmarịcha n'ígwé na-ejikwa udidi na obere nkọwa.
Kedu nsogbu StyleGAN2 doziri gbasara nke mbụ?
StyleGAN2 jiri demodulation arọ dochie AdaIN, wepụ droplet/blob artifacts na ịkwalite ogo onyonyo.