Igisekuru cya Autoregressive Igisekuru
Autoregressive image generation yubaka amashusho igice kimwe icyarimwe, guhanura buri kimenyetso mubintu byose byakozwe mbere yacyo.
Incamake
It matters because the same next-token machinery powering language models can produce coherent, controllable images.
Kwibira cyane
Autoregressive image generation ifata ifoto nkurukurikirane kandi ikanahanura ibice kubintu, aho buri kintu gishya giteganijwe kubibanjirije byose. Akazi kambere nka PixelRNN na PixelCNN yahanuye amashusho pigiseli mbisi icyarimwe, gusikana umurongo kumurongo, byatinze ariko bifite isuku mubyukuri. Sisitemu zigezweho ahubwo zabanje guhuza ishusho murusobe rwibimenyetso byihariye ukoresheje kodegisi ya VQ-VAE, hanyuma Transformer ihanura ibyo bimenyetso ibumoso-iburyo. __AIU_PROTECTED_10 __ 'DALL-E 1 na __AIU_PROTECTED_11 __' Parti yakurikije iyi resept, itanga ibimenyetso byamashusho byateganijwe kumyandiko mbere yo kubisubiza kuri pigiseli. Inyungu nini nuburyo bushoboka bwo kwerekana imiterere hamwe nubwubatsi bumwe busangiwe nururimi. Igiciro ni gikurikiranye, gahoro gahoro.
Ubushishozi
Icyitegererezo cyerekana uburyo bushoboka bwibimenyetso byose mubicuruzwa bisabwa: p (x) = ibicuruzwa bya p (x_i yahawe x_1 ... x_ {i-1}). Transformer hamwe nimpamvu (mask) kwitondera ishimangira ko buri mwanya ubona gusa ibimenyetso byambere. Mugihe cy'amahugurwa iteganya buri kimenyetso kibangikanye hakoreshejwe guhatira abarimu, ariko kubisobanuro bigomba kwerekana icyitegererezo kimwe icyarimwe, kugaburira buri inyuma.
Ingaruka z'Ingamba
Umuvuduko n'igipimo
AI igaragara irashobora gukora igenzura, gutahura, no gutondekanya imirimo kurwego.
Build choices
Amakipe arema arashobora prototype ibitekerezo byihuse hamwe nintoki nkeya.
Itsinda hamwe nakazi
Ibikorwa birashobora gukoresha amashusho nibimenyetso bya videwo byari bigoye gutunganya.
Ejo hazaza ha Autoregressive Ishusho Igisekuru
Umuvuduko nintambara yo hagati. Ubuhanga nka parallel hamwe na masked-token decoding (MaskGIT, Muse) ibyara ibimenyetso byinshi icyarimwe, kandi decoding yibitekerezo yatijwe mubyitegererezo byururimi irimo guhuzwa namashusho. Abashakashatsi kandi bahuza ibimenyetso nibishusho mumurongo umwe wa autoregressive umugongo kugirango moderi imwe ibashe gusoma no gushushanya, nkuko bigaragara muri sisitemu ya multimodal. Tegereza ibitekerezo bya autoregressive na diffuzion kugirango ukomeze kuvanga, hamwe na moderi ya Hybrid ifata kugenzura ibimenyetso nibiranga ikwirakwizwa.
Gushyira mu bikorwa Isi
DALL-E 1 yabyaye amashusho muburyo bwo guhanura urusobekerane rwibimenyetso byerekana amashusho uhereye kumyandiko.
__AIU_PROTECTED_11 __ 'Parti yagabanije inyandiko-yerekana-Guhindura inyandiko-shusho Guhindura kugeza kuri miliyari 20 ibipimo birambuye, byihuse-byizerwa.
PixelCNN na PixelRNN berekanye pigiseli mbisi-kuri-pigiseli kandi iracyakoreshwa nkibyingenzi byo kwigisha kubintu bishoboka.
MaskGIT na Muse bakoresha parallel ya mask-token decoding kugirango yihutishe icyerekezo gishingiye kumashusho mugihe ukomeje imyitozo ya autoregressive.
Ingaruka & Kurinda
Uburenganzira bwishusho hamwe no kwemererwa birashobora guhinduka ibyago byemewe n'amategeko niba ibimenyetso bidasobanutse.
Imikorere yicyitegererezo irashobora gutandukana kumurika, demografiya, nibidukikije.
Ibyiza byibinyoma birashobora kutamenyekana keretse niba ibyiringiro byateganijwe bikurikiranwa.
Igishushanyo mbonera
Sobanura ibipimo byo kwemererwa kugiciro, kwibutsa, nibiciro byamakosa.
Gerageza hamwe namakuru ajyanye nuburyo nyabwo bwo gukora.
Ongeraho isubiramo ryabantu kubwizere buke cyangwa guhanura cyane.
Kurikirana icyitegererezo cya drift hanyuma uhindurwe nyuma ya kamera cyangwa dataset ihinduka.
Komeza Ubushakashatsi
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 Autoregressive Image Generation 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
Ubuyobozi bukurikira
Igisekuru cya AI
Ibibazo bikunze kubazwa
What is Autoregressive Image Generation?
Autoregressive image generation yubaka amashusho igice kimwe icyarimwe, guhanura buri kimenyetso mubintu byose byakozwe mbere yacyo. Nibyiza kuko imashini imwe ikurikira-token imashini ikoresha imvugo yindimi irashobora gutanga amashusho afatika, agenzurwa.
'Autoregressive' bisobanura iki murwego rwo kubyara amashusho?
Moderi ya Autoregressive yerekana ishusho nkurukurikirane, guhanura buri kimenyetso cyangwa pigiseli ukurikije ibintu byose byakozwe mbere yacyo.
Nigute bigezweho bigezweho byerekana amashusho nka DALL-E 1 mubisanzwe byerekana ishusho mbere yo kubihanura?
Sisitemu zigezweho zifata ishusho mubimenyetso byihariye (akenshi binyuze kuri VQ-VAE-yuburyo bwa kodegisi), hanyuma uhanure icyo kimenyetso gikurikirana hamwe na Transformer.
Ni ubuhe bwoko bwambere bwatanze amashusho pigiseli mbisi icyarimwe?
PixelRNN na PixelCNN bari intangarugero ya autoregressive yerekana scanne kandi yahanuye amashusho pigiseli kuri pigiseli.
Nubuhe buryo butuma Transformer yitabira gusa ibimenyetso byambere mugihe cya autoregressive generation?
Impamvu yo guhisha ibuza buri mwanya kwitabira kwitabira ibimenyetso bizaza, bigashyira mubikorwa ibumoso-iburyo.
Ni izihe ngaruka nyamukuru zifatika zo gutoranya amashusho?
Kuberako buri kimenyetso giterwa nicyabanjirije, umwanzuro ugomba gukoresha ibimenyetso byikimenyetso, bigatuma igisekuru ugereranije buhoro.