Muse Masked Amashusho Yibyara
Muse ni inyandiko-y-ishusho yerekana kuva Google itanga amashusho wuzuza ibimenyetso byerekana amashusho icyarimwe icyarimwe, bigatuma byihuta cyane kuruta gukwirakwiza intambwe ku yindi.
Incamake
It matters because it showed you can get high-quality, well-aligned images without the slow iterative denoising that most generators rely on.
Kwibira cyane
Muse ikora mumwanya wihariye wa shusho. VQGAN yateguwe ihindura ishusho murusobe rwibimenyetso byuzuye, nkamagambo yamagambo yo kubaka. Mugihe cyamahugurwa, igice kinini cyibi bimenyetso cyarashyizwe ahagaragara, kandi Transformer yiga kubihanura inyuma, bigashyirwa kumyandiko yatanzwe kuva mururimi runini rwahagaritswe (T5-XXL). Mugihe cyibisekuru Muse atangirira kuri gride yose yipfundikiriye kandi ikanabangikanya mukuzenguruka, guhanura ibimenyetso byinshi kuri buri ntambwe no kongera guhisha ibyiringiro bike. Igishushanyo mbonera cyibice bibiri kibanza gutanga icyerekezo gike-token grid, hanyuma super-resolution-moderi yuzuza urwego rwo hejuru-rukuruzi. Kuberako ibimenyetso byinshi byakemuye icyarimwe, moderi ya 900M na 3B itanga ishusho ya pigiseli 256 cyangwa 512 mumashusho make yimbere.
Ubushishozi
Amayeri yibanze ni parallel decoding hamwe nicyizere gishingiye kuri remasking, bakunze kwita MaskGIT-yuburyo bw'icyitegererezo. Aho guhanura ikimenyetso kimwe icyarimwe (autoregressive) cyangwa kwerekana inshuro magana (diffusion), Musehanura ibimenyetso byose bipfutse mu maso, bikomeza ibyiringiro cyane, kandi byongeye guhisha ibisigaye mugice gikurikira. Gukoresha kodegisi ya T5-XXL yakonjeshejwe itanga ururimi rukomeye kubuntu, kandi gukora kubimenyetso byihariye bireka impamvu yicyitegererezo kubyerekeye amashusho cyane nkamagambo.
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 Muse Masked Yerekana Amashusho
Masked parallel decoding point yerekeza kuri generator zifite ubuziranenge kandi bwihuse rwose, nibyingenzi muguhindura hamwe no gukoresha ibikoresho. Tegereza igitekerezo-gihanura igitekerezo cyo guhuza hamwe no gukwirakwiza uburyo bwa videwo, hamwe nimbaraga zo guhita ushushanya, gusohora, hamwe no guhindura mask. Mugihe ibimenyetso bya disiketi bigenda bitera imbere, amashusho yerekana masike arashobora kwaguka neza muri videwo na 3D, aho decode ibangikanye ishobora kugabanya cyane ikiguzi cyo kubyara amakadiri menshi cyangwa ibitekerezo.
Gushyira mu bikorwa Isi
Ubuhanga bwihuse hamwe nibibaho aho umuhanzi akenera amashusho menshi mumasegonda kuruta iminota.
Gushushanya zeru-zeru, nko gukuraho ikintu no kugira icyitegererezo cyuzuza agace kegeranye gahoraho hamwe nibidukikije.
Igishushanyo cyo kwagura ifoto irenze imipaka yumwimerere kuri banneri cyangwa ibipimo bitandukanye.
Guhindura mask-yubusa, nko guhindura ibara ryimbwa cyangwa ikirere izuba rirenze muguhindura inyandiko hanyuma ukongera ugashiraho ibimenyetso byagize ingaruka.
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 Muse Masked Generative Imaging 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
Masked Autoencoders
Ibibazo bikunze kubazwa
What is Muse Masked Generative Imaging?
Muse ni inyandiko-y-ishusho yerekana kuva Google itanga amashusho wuzuza ibimenyetso byerekana amashusho icyarimwe icyarimwe, bigatuma byihuta cyane kuruta gukwirakwiza intambwe ku yindi. Nibyingenzi kuko byerekanaga ko ushobora kubona amashusho yujuje ubuziranenge, ahujwe neza nta gahoro gahoro gahoro yerekana ko generator nyinshi zishingiye.
Niki Musehanura mugihe cyibisekuru aho kwerekana pigiseli?
Muse ikorera mumwanya wihariye, uhanura ibimenyetso byerekana amashusho byakozwe na VQGAN tokenizer aho gukora kuri pigiseli itaziguye.
Ni ukubera iki muri rusange Muse yihuta kuruta uburyo busanzwe bwo gukwirakwiza?
Muse yuzuza ibimenyetso byinshi byapfunditswe icyarimwe kandi ikenera gusa uruziga rwa decoding, bitandukanye no gukwirakwiza intambwe nyinshi zikurikirana.
Muse yakura he gusobanukirwa ibyanditswe byihuse?
Muse imiterere yibisekuruza byanditse mubyanditswe kuva T5-XXL yerekana ururimi rwakonje, bikaruha kumva neza ururimi.
Ni uruhe ruhare rwo gusubiramo ibyiringiro muri Muse?
Nyuma ya buri kintu kibangikanye, Muse agumana ibimenyetso byizewe kandi yongera guhisha ibyiringiro bike kugirango atunganyirizwe hamwe.
Nigute Muse akora kubyara amashusho arenze?
Muse yabanje kubyara rezo-ntoya ya token ya gride, hanyuma moderi ya kabiri super-rezo yerekana ibyerekezo bihanitse.