Amabwiriza ya AI

Impinduka

Diffusion Transformers (DiTs) bahinduranya U-Net ihamye kumutima wibishusho na generator ya videwo kumugongo wa Transformer.

2 min somaIbiherutse kuvugururwa

Incamake

This architecture powers leading systems like Stable Diffusion 3 and OpenAI's Sora, and it scales remarkably well as you add compute.

Kwibira cyane

Moderi ya Diffusion itanga amashusho mugutangirira ku rusaku rwiza no kubisubiramo mubishusho bihuye. Kumyaka umuyoboro ukora iyo denoising yari U-Net, imyubakire yemeza. Impinduka ya Diffusion, yatangijwe na Peebles na Xie mu 2022, isimbuza U-Net na Transformer. Ishusho yabanje gukusanyirizwa mumwanya wihishe, igabanyijemo uduce duto, kandi buri patch iba ikimenyetso, cyane nkamagambo muburyo bw'ururimi. Transformer noneho itunganya ibi bimenyetso hamwe no kwiyitaho kuri buri ntambwe yerekana. Icy'ingenzi cyagaragaye ni uko imikorere ya DiT itera imbere mu buryo buteganijwe uko wongera ubunini bw'icyitegererezo kandi ukagabanya ingano ya patch, ukurikiza amategeko yo gupima neza. Ubu bunini ni impamvu sisitemu-kuri-videwo na sisitemu yohejuru-yinyandiko-y-amashusho yimukiye ahanini muri Transformer umugongo.

Ubushishozi

Guhanga udushya nuburyo DiTs itera inshinge nka timestep hamwe ninyandiko yihuse. Aho guhuriza hamwe byoroshye, bakoresha imiterere ihuza imiterere (adaLN), aho umuyoboro uteganya ibipimo noguhindura ibipimo bisanzwe mubisanzwe uhereye kubimenyetso byerekana. Impinduka ya adaLN-zeru itangiza ibi rero buri gice gitangira nkigikorwa cyiranga, gihagarika amahugurwa. Ibishishwa bisobekeranye mubimenyetso, bitunganywa nibisanzwe bya Transformer hamwe no kwiyitaho, hanyuma bigateranyirizwa hamwe hanyuma bigasubira muri pigiseli.

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 h'abahindura Diffusion

Impinduka za Diffusion ziba umugongo wambere kubitangazamakuru bitanga umusaruro. Igishushanyo cyabo gishingiye kubimenyetso bituma bakora muburyo bwo guhuza amashusho, videwo, ndetse nibisekuru byinshi munsi yububiko bumwe. Ubushakashatsi burimo gusunika kuri videwo ndende, gukemura cyane, no kurushaho kwitabwaho kugirango ugabanye igiciro cya kane cyibimenyetso byinshi. Tegereza guhuza imvugo nicyitegererezo, aho bisa na Transformer scaring resept hamwe nibikorwa remezo bikora byombi, byihutisha iterambere mubyitegererezo byisi na videwo ikorana.

Gushyira mu bikorwa Isi

__AIU_PROTECTED_10 __ 'Sora ikoresha umugongo wa Transformer hejuru yigihe cyagenwe kugirango ubyare amashusho yiminota-minini, yizerwa cyane uhereye kumyandiko.

Diffusion Stable 3 ifata Multimodal Diffusion Transformer (MMDiT) kugirango ihuze neza amashusho yakozwe hamwe nibisobanuro birambuye byanditse.

Abashakashatsi bapima DiT kugeza kuri miliyari y'ibipimo kandi bakareba ubwiza bw'amashusho bugenda butera imbere, bayobora ibyemezo bya bije.

Sitidiyo ikoresha moderi ishingiye kuri DiT kugirango yongere amashusho magufi, ifata amashusho yandi mashusho nkibimenyetso byinyongera byerekana.

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

1

Sobanura ibipimo byo kwemererwa kugiciro, kwibutsa, nibiciro byamakosa.

2

Gerageza hamwe namakuru ajyanye nuburyo nyabwo bwo gukora.

3

Ongeraho isubiramo ryabantu kubwizere buke cyangwa guhanura cyane.

4

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 Diffusion Transformers quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Tangira ikibazo

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Ubuyobozi bukurikira

Imiyoboro ihinduranya

Ibibazo bikunze kubazwa

What is Diffusion Transformers?

Diffusion Transformers (DiTs) bahinduranya U-Net ihamye kumutima wibishusho na generator ya videwo kumugongo wa Transformer. Ubu bwubatsi bufite imbaraga ziyobora sisitemu nka Stable Diffusion 3 na __AIU_PROTECTED_10 __ 'Sora, kandi ipima neza kuburyo wongeyeho compte.

Niki kintu gihindura Diffusion Transformer ugereranije na moderi gakondo yo gukwirakwiza?

DiTs isimbuza U-Net, umuyoboro wemeza wakoze denoising, hamwe numugongo wa Transformer.

Nigute DiT ihindura ishusho mubintu Transformer ishobora gutunganya?

Ishusho yihishe igabanijwemo uduce duto, kandi buri patch iba ikimenyetso, gisa namagambo murugero rwururimi.

Ni iki impapuro z'umwimerere DiT zabonye zijyanye no gupima?

Ubwiza bwa DiT butera imbere muburyo busukuye, buteganijwe uko wongera compte yicyitegererezo kandi ugakoresha uduce duto, ukurikiza amategeko yo gupima.

Nigute DiTs itera inshinge nka timestep hamwe nibisobanuro byihuse?

DiTs ikoresha uburyo bwo guhuza n'imihindagurikire y'ikirere kugira ngo hamenyekane ibipimo n'ibipimo byerekana ibipimo bisanzwe biva ku kimenyetso cyerekana.

Ni iki gutangiza 'adaLN-zeru' bigeraho?

adaLN-zeru itangiza modulation kuburyo buri gice kibanza gukora nkirangamuntu, ituza kandi igateza imbere amahugurwa.