I-Make-A-Video Text-to-Video
I-Make-A-Video iwuhlelo luka-Meta luka-2022 olushintsha ukwaziswa kombhalo ube isiqeshana sevidiyo esifushane ngaphandle kokuqeqeshwa ngamapheya anelebula yombhalo-vidiyo.
Uhlolojikelele
It matters because it showed that the visual knowledge inside text-to-image models could be 'taught' to move using only unlabeled video.
I-Deep Dive
I-Make-A-Video, imenyezelwe yi-Meta AI ngo-Septhemba 2022, ikhiqiza imizuzwana embalwa yevidiyo emshweni ofana 'nenja egqoke ikepisi leqhawe elindiza esibhakabhakeni.' Iqhinga layo eliyinhloko ukuhlukanisa ukubukeka kusuka ekunyakazeni: imodeli yombhalo uye esithombeni (eyakhelwe esikhaleni sesithombe esihlanganyelwe sesitayela se-CLIP) ifunda ukuthi izinto zibukeka kanjani ezigidini zezithombe ezinamagama-ncazo, kuyilapho izendlalelo ze-spatiotemporal ezihlukene zifunda ukuthi izinto zihamba kanjani kuvidiyo engenamalebula iyodwa. Lokhu kubeka eceleni ukushoda kwamapheya evidiyo yombhalo wekhwalithi ephezulu. Imodeli eyisisekelo ikhiqiza iziqeshana ezinokulungiswa okuphansi, ezinesilinganiso esiphansi, bese amanethiwekhi azinikele ahlanganisa ozimele abengeziwe kanye nokulungiswa kwendawo ephezulu. Umphumela ubuhambisana ngokumangalisayo ngesikhathi sawo, nakuba iziqeshana zazimfushane, zifiphele, futhi zijwayele ukucwayiza futhi zinyakaze.
I-Technical Insight
I-Make-A-Video inweba ukuguqulwa kwesizukulwane sesithombe se-2D nokunaka ku-3D ngokungeza izendlalelo zesikhashana-mbumbulu. Izisindo zendawo eziqeqeshelwe kusengaphambili ziyaqandiswa noma zicushwe kahle kuyilapho izendlalelo zesikhashana ezintsha zifunda ukunyakaza kuvidiyo eluhlaza, ngakho awekho amalebula evidiyo yombhalo adingekayo. Inethiwekhi yokuhumusha kozimele ibe isigxilisa umugqa wesikhathi kanye namamojula okusabalalisa okucacayo okuphezulu aphakamisa imininingwane yendawo, aguqule uhlaka olumahhadla-16, olunokukhanya okuphansi lube isiqeshana esibushelelezi, esibukhali epayipini elijikijelayo.
I-Strategic Impact
Isivinini nesikali
I-Visual AI ingakwazi ukuhlola, ukutholwa, nokumaka imisebenzi esikalini.
Yakha ukukhetha
Amathimba aqanjiwe angakwazi ukulinganisa imiqondo ngokushesha ngezibuyekezo ezimbalwa ezenziwa mathupha.
Ithimba kanye nokusebenza komsebenzi
Imisebenzi ingasebenzisa amasiginali wesithombe nawevidiyo obekunzima ukuwenza ngaphambilini.
Ikusasa le-Make-A-Video Text-to-Video
Isithombe se-Make-A-Video's image-prior-plus-unlabel-motion recipe sikhiqize wonke amagagasi ombhalo ukuya kuvidiyo. Inzalo yayo igcizelela iziqeshana ezinde, ezinokulungiswa okuphezulu, ezinzile okwesikhashana ezinokunyakaza kwekhamera elawulekayo nomsindo. Lindela umqondo oyinhloko, ukusebenzisa kabusha ulwazi olukhulu lwesithombe nokufunda ukunyakaza okushibhile, ukuqhubeka ngisho noma izakhiwo zishintshela ekusakazeni okucashile okusekelwe ku-transformer kanye namamodeli ahlanganisiwe amukela isimo sesithombe noma sevidiyo ukuze sihlelwe futhi siqhubeke.
Ukuqaliswa Komhlaba Wangempela
Ukugqwayiza umusho owodwa ochazayo ube isiqeshana esifushane sokuthunyelwe kwenkundla yezokuxhumana
Ukuletha umqondo omile njengokuthi 'ibhere lika-teddy lipenda isithombe' njengomfanekiso onyakazayo
Ukuhunyushwa phakathi kwezithombe ezimbili ezimile ezihlinzekwe ngabasebenzisi ukudala ividiyo yoguquko ebushelelezi
Ikhiqiza okusalungiswa okunyakazayo okusheshayo kwezigcawu ezicatshangelwayo zokuqoshwa kwendaba ngaphambi kwanoma yikuphi ukuqoshwa
Izingozi & Guardrails
Amalungelo ezithombe kanye nemvume kungaba ubungozi bezomthetho uma ukuvela kungacacile.
Ukusebenza kwemodeli kungahluka kukho konke ukukhanya, izibalo zabantu, kanye nezindawo.
Okuhle okungelona iqiniso kungase kungabonakali ngaphandle uma izinga lokuzethemba liqashelwa.
Ukuqalisa Umhlahlandlela
Chaza indlela yokwamukela yokunemba, ukukhumbula, nezindleko zamaphutha.
Hlola ngedatha efana nezimo zangempela zokukhiqiza.
Engeza isibuyekezo somuntu ukuze uthole ukuzethemba okuphansi noma izibikezelo zomthelela omkhulu.
Landelela ukukhukhuleka kwemodeli bese uqinisekisa kabusha ngemva kwezinguquko zekhamera noma zesethi yedatha.
Qhubeka Uhlole
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 Make-A-Video Text-to-Video 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
Umhlahlandlela olandelayo
Sora kanye nokuguqula umbhalo ube yividiyo
Imibuzo evame ukubuzwa
What is Make-A-Video Text-to-Video?
I-Make-A-Video iwuhlelo luka-Meta luka-2022 olushintsha ukwaziswa kombhalo ube isiqeshana sevidiyo esifushane ngaphandle kokuqeqeshwa ngamapheya anelebula yombhalo-vidiyo. Kubalulekile ngoba kubonise ukuthi ulwazi olubonakalayo ngaphakathi kwamamodeli wombhalo uye esithombeni 'lungafundiswa' ukunyakaza kusetshenziswa ividiyo engenamalebula kuphela.
Iyiphi indlela entsha eqanjiwe evumela i-Make-A-Video ukuthi igweme ukudinga amapheya anelebula yombhalo-vidiyo?
I-Make-A-Video iqeda inkinga: imodeli yombhalo uye esithombeni ifunda ukuthi izinto zibukeka kanjani ezithombeni ezinamagama-ncazo, kuyilapho izendlalelo zesikhashana ezihlukene zifunda ukunyakaza kuvidiyo eluhlaza, engenamalebula.
I-Make-A-Video iwangeza kanjani amandla okunyakaza kumodeli yesithombe?
Inweba ukuguquguquka kwesikhala kwe-2D nokunaka ngezendlalelo zesikhashana ezintsha ezifunda ukuthi okuqukethwe kushintsha kanjani ngokuhamba kwesikhathi.
Zenzani izigaba ze-frame-interpolation kanye ne-super-resolution?
Ngemuva kokusalungiswa okuphansi kwamaholo, uhlaka oluphansi, ukuhumusha kwengeza amafreyimu futhi amamojula anokulungiswa okuphezulu aphakamisa ukulungiswa.
Wawuyini umkhawulo ojwayelekile wokuphumayo kwe-Make-A-Video?
Iziqeshana zaziyimizuzwana embalwa kuphela, ukulungiswa okuphansi uma kuqhathaniswa, futhi zijwayele ukucwayiza nokuhlanekezeka kwesikhashana.