Ọgbọ Onyonyo Autoregressive
Ọgbọ onyonyo autoregressive na-ewulite foto otu ibe n'otu oge, na-ebu amụma nke ọ bụla sitere na ihe niile emepụtara n'ihu ya.
Nchịkọta
It matters because the same next-token machinery powering language models can produce coherent, controllable images.
Ime miri emi
Ọgbọ onyonyo autoregressive na-ele foto anya dị ka usoro wee buru amụma ya site na mmewere, ebe ihe ọhụrụ ọ bụla na-adabara na ndị gara aga. Ọrụ mbụ dị ka PixelRNN na PixelCNN buru amụma onyonyo otu pikselụ raw n'otu oge, na-enyocha ahịrị n'usoro, nke na-adị nwayọ mana n'ụzọ doro anya. Sistemu ọgbara ọhụrụ kama buru ụzọ pịnye onyonyo n'ime grid nke akara ngosi dị iche iche site na iji koodu ntinye ụdị VQ-VAE, mgbe ahụ, Transformer na-ebu amụma akara ngosi ndị ahụ aka ekpe gaa n'aka nri. OpenAI's DALL-E 1 na Google's Parti gbasoro uzommeputa a, na-ewepụta akara ngosi onyonyo dabara na ederede ozugbo tupu ịmegharịa ha na pikselụ. Nnukwu uru bụ ihe nlere anya nke ọma yana ụlọ ejikọtara ọnụ na asụsụ. Ọnụ ego a na-aga n'usoro, nlele ngwa ngwa.
Nghọta nka nka
Ihe nlere anya na-eme ka njikọ nke akara niile bụrụ ngwaahịa nke ọnọdụ: p(x) = ngwaahịa nke p(x_i nyere x_1...x_{i-1}). Onye ntụgharị nwere nlebara anya ihe kpatara (ihe mkpuchi) na-amanye na ọnọdụ ọ bụla na-ahụ naanị akara ngosi mbụ. N'oge ọzụzụ, ọ na-ebu amụma maka akara ọ bụla n'otu n'otu site na iji mmanye onye nkuzi, mana n'echiche, ọ ga-enwerịrị nlele otu akara n'otu oge, na-azụ onye ọ bụla azụ azụ. Akwụkwọ koodu a mụtara na-egosi akara azụ na patches onyonyo, nke decoder na-ebuli elu n'ime pikselụ ikpeazụ.
Mmetụta atụmatụ
Ọsọ na ọnụ ọgụgụ
Visual AI nwere ike megharịa nyocha, nchọpụta na mkpado ọrụ n'ọtụtụ.
Mee nhọrọ
Otu ndị na-emepụta ihe nwere ike imepụta echiche ngwa ngwa site na ngbanwe akwụkwọ ntuziaka ole na ole.
Team na usoro ọrụ
Ọrụ nwere ike iji onyonyo na akara vidiyo siri ike ịhazi.
Ọdịnihu nke Ọgbọ onyonyo Autoregressive
Ọsọ bụ ebe a na-alụ ọgụ. Usoro dị ka myirịta na ihe nkpuchi nkpuchi (MaskGIT, Muse) na-ewepụta ọtụtụ akara n'otu oge, na ngbanwe nke agbaziri site na ụdị asụsụ na-eme ka ọ bụrụ onyonyo. Ndị na-eme nchọpụta na-ejikọta ederede na ihe oyiyi oyiyi n'otu ọkpụkpụ azụ azụ ka otu ụdị nwee ike ịgụ na ise, dị ka a na-ahụ na usoro multimodal. Na-atụ anya na echiche autoregressive na mgbasa ozi ga-anọgide na-agwakọta, yana ụdị ngwakọ na-ejide njikwa akara na ogo mgbasa ozi.
Mmejuputa n'ezie n'ụwa
DALL-E 1 weputara onyogho site n'ịtụgharị aka na-ebu amụma grid nke akara ngosi onyonyo dị iche site na ntinye ederede.
Google's Parti webatara ntụgharị ntụgharị ederede gaa na onyonyo gaa na ijeri 20 maka nkọwa zuru ezu, ihe ngosi kwesịrị ntụkwasị obi ozugbo.
PixelCNN na PixelRNN gosipụtara ọgbọ pikselụ-site-pixel raw ma ka na-ejikwa ya dị ka usoro nkuzi maka ụdị dabere na ohere.
MaskGIT na Muse na-eji ngbanwe ihe nkpuchi ihe mkpuchi na-emekọ ihe iji mee ka njikọ onyonyo dabere na ngwa ngwa ka ha na-echekwa ọzụzụ n'ụdị autoregressive.
Ihe ize ndụ & okporo ụzọ nche
Ikike onyonyo na nkwenye nwere ike bụrụ ihe egwu dị n'iwu ma ọ bụrụ na edoghị anya.
Ọrụ nlereanya nwere ike ịdịgasị iche n'ofe ọkụ, igwe mmadụ, na gburugburu.
Enwere ike ghara ịhụ ihe dị mma ma ọ bụrụ na enyochaghị oke ntụkwasị obi.
Map mmejuputa
Kọwaa ụkpụrụ nnabata maka nkenke, icheta, na ụgwọ njehie.
Nwalee na data dabara na ọnọdụ mmepụta n'ezie.
Tinye nyocha mmadụ maka obere obi ike ma ọ bụ amụma mmetụta dị elu.
Sochie ihe nlere anya wee megharịa ka emechara mgbanwe igwefoto ma ọ bụ dataset.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Ọgbọ onyonyo AI
Ajụjụ a na-ajụkarị
What is Autoregressive Image Generation?
Ọgbọ onyonyo autoregressive na-ewulite foto otu ibe n'otu oge, na-ebu amụma nke ọ bụla sitere na ihe niile emepụtara n'ihu ya. Ọ dị mkpa n'ihi na otu igwe na-esote akara asụsụ na-enye ụdị asụsụ nwere ike iwepụta onyonyo ejikọta ọnụ, enwere ike ịchịkwa.
Kedu ihe 'autoregressive' pụtara n'ihe gbasara ọgbọ onyonyo?
Modelsdị autoregressive na-emepụta onyonyo dị ka usoro, na-ebu amụma akara ma ọ bụ pikselụ ọ bụla dabere na ihe niile ewepụtara n'ihu ya.
Kedu ka ụdị onyonyo autoregressive ọgbara ọhụrụ dị ka DALL-E 1 si anọchi anya onyonyo tupu ibu amụma ya?
Sistemụ nke ọgbara ọhụrụ na-akpakọ onyonyo a ka ọ bụrụ akara ngosi pụrụ iche (na-abụkarị site na ngbanwe ụdị VQ-VAE), wee buru amụma usoro akara ahụ site na iji Transformer.
Kedu ụdị mbụ mepụtara onyonyo otu pikselụ raw n'otu oge?
PixelRNN na PixelCNN bụ ndị na-asụ ụzọ na ụdị autoregressive na-enyocha ma buru amụma pixel site na pikselụ.
Kedu usoro na-eme ka onye ntụgharị na-aga naanị akara ngosi mbụ n'oge ọgbọ na-agba ọsọ?
Mkpuchi ihe kpatara na-egbochi ọnọdụ ọ bụla ịga na akara ngosi n'ọdịnihu, na-eme ka ikike aka ekpe gaa aka nri.
Gịnị bụ isi ihe ndọghachi azụ nke autoregressive image sampling?
N'ihi na akara nke ọ bụla na-adabere na nke ndị gara aga, ntinye akwụkwọ ga-eji akara na-eme ka ọgbọ na-adị nwayọọ.