Table of Contents[Hide][Show]
Qhov tshiab thiab txhim kho AI tau txhim kho lub peev xwm, kev nkag siab, thiab lub peev xwm los tsim cov duab zoo dua. Tej zaum koj yuav tsis ntev los no tuaj hla qee cov duab coj txawv txawv thiab lom zem ntab ncig hauv internet.
Ib tug dev Shiba Inu hnav khaub ncaws nyob rau hauv ib tug beret thiab ib tug dub turtleneck. Thiab ib tug hiav txwv otter raws li Dutch neeg pleev kob Vermeer's "Girl with a Pearl Earring." Thiab muaj ib khob kua zaub uas zoo li tus dab woolly.
Cov duab no tsis yog tsim los ntawm tib neeg artist.
Hloov chaw, DALL-E 2, AI system tshiab uas tuaj yeem hloov cov lus piav qhia rau hauv cov duab, tsim lawv.
Tsuas yog sau dab tsi koj xav pom, thiab AI yuav tsim nws rau koj - hauv kev nthuav dav, zoo heev, thiab, qee zaum, kev tsim kho tiag tiag. Hauv tsab xov xwm no, peb yuav ua tib zoo saib ntawm OpenAI txoj kev tshawb fawb tshiab, DALL.E 2, nrog rau nws ua haujlwm li cas, thiab ntau ntxiv. Cia peb pib.
Yog li ntawd, yog dab tsi tiag DALL.E 2?
DALL-E 2 yog "tus qauv tsim," ib hom kev kawm tshuab algorithm uas tsim cov khoom siv nyuaj es tsis ua qhov kev kwv yees lossis kev faib ua haujlwm ntawm cov ntaub ntawv nkag.
Koj muab DALL-E 2 nrog cov lus piav qhia, thiab nws tsim ib daim duab uas sib haum rau nws. Los ntawm kev sib txuas cov ntsiab lus, qhov zoo, thiab cov qauv, OpenAI's DALLE 2 tuaj yeem tsim cov duab tshiab, muaj tseeb thiab kos duab los ntawm cov lus piav qhia yooj yim.
Qhov tseeb version, DALLE 2, tau hais tias muaj ntau yam, muaj peev xwm ua cov duab los ntawm cov ntawv sau npe ntawm cov kev daws teeb meem siab dua thiab hauv qhov dav dav ntawm cov qauv tsim. Piv txwv li, cov duab hauv qab no (los ntawm DALL-E 2 blog post) yog tsim los ntawm cov lus piav qhia "Ib tug astronaut caij nees."
Ib qho kev piav qhia xaus, "zoo li tus cwj mem kos duab," thaum lwm qhov xaus, "zoo li photorealistic."
Nws kuj tseem tuaj yeem hloov cov duab uas twb muaj lawm nrog kev xav tsis thoob. Yog li, koj tuaj yeem ntxiv lossis tshem tawm cov ntsiab lus thaum khaws cov xim, kev xav, thiab duab ntxoov ntxoo, tag nrho thaum tuav cov duab qub.
Ua li cas nws ua hauj lwm?
DALL-E 2 ua rau siv CLIP thiab diffusion qauv, ob qho tib si sophisticated kawm tob txoj hauv kev tau tsim nyob rau xyoo tas los no. Txawm li cas los xij, nws yog raws li kev xav zoo ib yam li txhua qhov sib sib zog nqus Neural networks: kev kawm sawv cev. CLIP ib txhij qhia ob neural networks ntawm cov duab thiab kab lus.
Ib lub network kawm cov duab sawv cev hauv daim duab, thaum lwm tus kawm cov ntawv sawv cev. Thaum lub sij hawm kev cob qhia, ob lub tes hauj lwm sim hloov kho lawv qhov tsis sib xws kom cov duab sib piv thiab cov lus piav qhia ua rau muaj qhov sib xws.
"Diffusion," ib hom qauv tsim uas kawm ua cov duab los ntawm maj mam nrov nrov thiab denoising nws cov qauv kev cob qhia, yog lwm lub tshuab kev kawm siv hauv DALL-E 2. Diffusion qauv zoo ib yam li autoencoders uas lawv hloov cov ntaub ntawv nkag mus rau hauv ib qho embedding sawv cev thiab tom qab ntawd siv cov ntaub ntawv embedding los tsim cov ntaub ntawv qub.
Siv OpenAI's qauv lus CLIP, uas tuaj yeem txuas cov lus piav qhia nrog cov duab, nws thawj zaug txhais cov ntawv sau tam sim ntawd rau hauv daim ntawv nruab nrab uas suav nrog cov khoom tseem ceeb uas daim duab yuav tsum tau ua kom zoo li ntawd (raws li CLIP).
Qhov thib ob, DALL-E 2 tsim ib qho CLIP-raws li daim duab siv tus qauv diffusion, uas yog neural network.
Ntawm cov duab distorted nrog random pixels, diffusion qauv tau kawm. Lawv kawm yuav ua li cas los kho cov duab 'thawj daim ntawv. Cov qauv diffusion tuaj yeem tsim cov duab zoo nkauj zoo nkauj, tshwj xeeb tshaj yog thaum siv nrog rau kev coj ua kom pom qhov tseem ceeb ntawm qhov tseeb ntawm ntau haiv neeg.
Raws li qhov tshwm sim, tus qauv diffusion siv cov pixels random thiab siv CLIP los hloov lawv mus rau hauv cov duab tshiab uas phim cov lus qhia. Vim yog lub tswv yim diffusion, DALL-E 2 tuaj yeem tsim cov duab daws teeb meem ntau dua li DALL-E.
DALL.E 2 siv cov ntaub ntawv
Hauv nees nkaum xyoo dhau los, lub computer tsis pom kev thev naus laus zis tau nce qib los ntawm kev xav yooj yim mus rau qhov kev ua tiav loj. Txawm hais tias cov kev nce qib no, cov duab thiab cov qauv paub txog cov khoom tseem ntsib teeb meem tseem ceeb hauv lub neej txhua hnub. Qhov tsis muaj datasets yog ib qho teeb meem tseem ceeb tshaj plaws ntawm cov duab pom thiab pom lub computer. Vim tias muaj cov ntaub ntawv tsis txaus ntawm ob qho kawg, kev cob qhia cov duab paub txog cov qauv kom muab 100 feem pua tau raug yog yuav luag nyuaj.
Hmoov zoo, OpenAI tus qauv kawm tshuab tshiab tuaj yeem txuas qhov sib txawv hauv kev siv tshuab. DALLE 2 muaj peev xwm tsim cov duab zoo nkauj raws li cov lus piav qhia. Qhov kev tsim duab cuav no tuaj yeem muab cov ntaub ntawv rau cov duab pom cov qauv raws li lawv cov kev xav tau. Qhov tsis muaj cov ntaub ntawv yog qhov cuam tshuam tseem ceeb rau kev txheeb xyuas cov khoom thiab cov duab.
Hauv lub sijhawm digital, cov ntaub ntawv muaj nyob txhua qhov chaw, tab sis peb tseem tab tom nrhiav kev luv los pub AI qauv, yog li nws tuaj yeem muab cov txiaj ntsig zoo. Txawm li cas los xij, nws tsis yooj yim rau kev cob qhia tus qauv paub txog duab. Nws xav tau ntau cov ntaub ntawv teev nrog qhov sib txawv me me, uas peb yuav tsis tuaj yeem khaws tau yooj yim.
Yog li, lo lus teb yog dab tsi: Cov lus teb yog DALLE 2. Lub tshuab hluav taws xob OpenAI duab, nrog nws lub peev xwm los tsim cov duab los ntawm cov ntawv thiab hloov cov uas twb muaj lawm, tuaj yeem pab txuas qhov sib txawv. Qhov no yuav pab tsim cov ntaub ntawv kev cob qhia ntxiv thaum tseem txo cov neeg sau npe yuav tsum tau ua. Txawm tias muaj txiaj ntsig zoo, koj yuav tsum paub txog kev dag ntxias cov duab tsim tawm thiab cov duab uas tsis suav nrog suav nrog. Qhov no tuaj yeem ua rau cov duab tshawb pom cov txheej txheem tsim cov txiaj ntsig tsis ncaj ncees.
Cov kev txwv
DALL.E 2 tej zaum yuav muaj kev phom sij zoo yog tias nws poob rau hauv tes tsis raug, raws li OpenAI. Hauv lub ntiaj teb niaj hnub no ntawm kev dag ntxias sib sib zog nqus, tus qauv tuaj yeem siv tau yooj yim los nthuav tawm cov ntaub ntawv tsis tseeb lossis kev ntxub ntxaug lwm tus, uas yog vim li cas OpenAI tsuas tso cai rau cov neeg tsim khoom siv DALL.2 los ntawm kev caw. Tus qauv yuav tsum ua raws li kev txwv nruj rau txhua cov lus qhia uas nws tau txais.
Txhawm rau tshem tawm lub peev xwm ntawm DALL.E 2 tsim cov duab ua phem lossis ua phem, cov ntaub ntawv tau tsim tsis muaj riam phom tua neeg. Thaum OpenAI tau hais tias nws npaj yuav hloov nws mus rau API yav tom ntej, nyob rau hauv rooj plaub ntawm DALL.E 2, nws txaus siab ua kom ceev faj.
xaus
DALL-E 2 yog lwm qhov kev tshawb nrhiav OpenAI nthuav dav uas qhib qhov rooj rau cov ntawv thov tshiab.
Ib qho piv txwv yog tsim cov ntaub ntawv loj heev kom tau raws li ib qho ntawm lub khoos phis tawj lub zeem muag lub ntsiab fwj-cov ntaub ntawv. Thaum cov ntaub ntawv nyiaj txiag rau ntau DALL-E-raws li cov apps yuav raug txiav txim siab los ntawm tus nqi thiab cov cai uas OpenAI tsim rau nws cov neeg siv API, lawv txhua tus yuav tsis xav txog kev tsim cov duab.
Sau ntawv cia Ncua