Ụlọ ọrụ Wasserstein GAN
Wasserstein GAN (WGAN) bụ nhazigharị nke ebumnuche ọzụzụ GAN nke na-eji anya Wasserstein kama mfu min-max mbụ.
Nchịkọta
It makes notoriously unstable GAN training far more reliable and gives a loss value that actually correlates with image quality.
Ime miri emi
Ndị GAN mbụ na-azụ netwọkụ abụọ n'ọgụ ọgụ: onye na-emepụta ihe na-eme onyonyo adịgboroja na onye ịkpa oke na-anwa ịhụ ha. Nke a na-adakarị ma ọ bụ kwụsị n'ihi na mfu nke onye ịkpa ókè ekwughị ihe ọ bụla bara uru maka ọganihu. WGAN, nke Arjovsky, Chintala, na Bottou webatara na 2017, dochie onye ịkpa ókè na 'onye nkatọ' nke na-egosi etu ihe onyonyo si ele anya n'ọ̀tụ̀tụ̀ na-aga n'ihu kama ịkọwapụta ezigbo vs-adịgboroja. Ebumnuche ọzụzụ na-aghọ ebe dị anya nke Wasserstein (ụwa) n'etiti nkesa data n'ezie na ewepụtara. Ebe dị anya a na-enye gradients dị nro karị, bara uru ọbụna mgbe nkesa abụọ ahụ na-agbakọ, na-ebelata ndakpọ ọnọdụ n'ụzọ dị egwu ma na-eme ka mfu ahụ bụrụ akara ngosi dị mma.
Nghọta nka nka
Ebe dị anya nke Wasserstein na-eji nlezianya tụọ 'ọrụ' kacha nta iji mee ka otu ikpo unyi (nkesa adịgboroja) banye ọzọ (nke bụ nke bụ eziokwu). Kọmputa ya na-adabere na duality Kantorovich-Rubinstein, nke chọrọ ka onye nkatọ bụrụ 1-Lipschitz (gradients nwere oke). WGAN mbụ kwadoro nke a nke ọma site n'ibibi arọ gaa n'obere oke; WGAN-GP mechara jiri ntaramahụhụ gradient dochie clipping nke na-akwalite ụkpụrụ gradient nke onye nkatọ na 1, na-azụkwu nke ọma.
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 Wasserstein GAN
Nghọta bụ isi nke WGAN, na nhọrọ nkesa anya na-akpụzi ịdịmma gradient, ka na-ekwughachi site na imepụta ụdị. Ọ bụ ezie na ụdị mgbasa ozi ugbu a na-achịkwa njikọ onyonyo, echiche mbufe kacha mma sitere na WGAN na-apụtaghachi na ndakọrịta, ụzọ Schrodinger-bridge, na ịgbasa ụdị mgbasa n'ime ndị na-emepụta nzọụkwụ ole na ole ngwa ngwa. Na-atụ anya ebumnobi ụdị Wasserstein ka ọ nọgide na-agwa usoro ngwakọ ebe ọzụzụ kwụsiri ike na ihe metric mfu bara uru, ọkachasị na ngalaba sayensị na obere data.
Mmejuputa n'ezie n'ụwa
Na-emepụta ihu na textures dị adị ebe vanilla GAN dara na mpụta ole na ole ugboro ugboro
Ịmepụta onyonyo ahụike sịntetị, dị ka MRI ma ọ bụ patches histology, iji kwalite ihe ndekọ data akpọrọ ụkọ.
Ịmepụta ihe omume ndakọrịta n'ime ihe ngosi physics nwere ike dị elu ebe ọzụzụ kwụsiri ike dị oke mkpa
Na-eje ozi dị ka akara nrịbama n'ime nyocha ML n'ihi na ọnwụ ya na-esopụta ogo nlele karịa ọzụzụ
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 Wasserstein GAN 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
ESRGAN na GAN Super-Mkpebi
Ajụjụ a na-ajụkarị
What is Wasserstein GAN?
Wasserstein GAN (WGAN) bụ nhazigharị nke ebumnuche ọzụzụ GAN nke na-eji anya Wasserstein kama mfu min-max mbụ. Ọ na-eme ka ọzụzụ GAN a ma ama na-akwụsighị ike karịa nke a pụrụ ịdabere na ya ma na-enye uru mfu nke dabara na ogo onyonyo.
Kedu metric anya ka WGAN na-eji atụnyere nkesa n'ezie na ewepụtara?
WGAN jiri anya Wasserstein dochie ebumnuche mbụ dabere na Jensen-Shannon, nke na-enye gradients dị nro ọbụlagodi mgbe nkesa anaghị agafe.
Na WGAN, netwọk nke bụ onye ịkpa oke ka a kpọgharịrị aha ka ọ bụrụ gịnị, n'ihi gịnịkwa?
Onye nkatọ na-enyocha onyonyo n'ogo na-aga n'ihu kama ịkọwapụta ezigbo vs adịgboroja, nke bụ ihe na-eme ebumnuche Wasserstein.
Gịnị kpatara na a ga-amachibido onye nkatọ WGAN ịbụ 1-Lipschitz?
Ihe abụọ na-eme ka WGAN tụọ anya Wasserstein na-ejide naanị maka ọrụ 1-Lipschitz, yabụ na gradients nke onye nkatọ ga-enwerịrị oke.
Kedu ka WGAN mbụ siri kwado mmachi Lipschitz?
Akwụkwọ WGAN nke mbụ jiri mbelata arọ crude; WGAN-GP mechara jiri ntaramahụhụ gradient dị nro dochie ya.
Kedu nsogbu WGAN na-ebelata nke ọma ma e jiri ya tụnyere vanilla GANs?
WGAN's smoother gradients na-egbochi ọnọdụ ndakpọ wee mepụta mfu nke dabara na ogo nlele.