Overview
MaskGIT inogadzira mifananidzo nekufanotaura zviratidzo zvakawanda kamwechete uye kuzadza kune vane chivimbo zvakanyanya kutanga, kutsiva chizvarwa chinononoka kubva kuruboshwe kuenda-kurudyi nematanho mashoma anoenderana.
MaskGIT Parallel Token Decoding ndeyekombuta-kuona mafambiro anodudzira kana kugadzira midhiya yekuona yekuongorora, mashandiro, uye kugadzira.
Deep Dive
MaskGIT (Masked Generative Image Transformer), kubva Google muna 2022, inofunga zvakare magadzirirwo emhando yemifananidzo. Kare vashanduri vakaita seVQGAN vakagadzira tokens autoregressively, imwe panguva mune raster order, inononoka uye isiri yechisikigo yemifananidzo ye2D. MaskGIT pachinzvimbo inodzidzisa ine masked modelling vavariro seBERT: zvisina tsarukano subsets yemifananidzo tokens yakavanzwa uye modhi inodzidza kufanotaura zvese panguva imwe chete uchishandisa bidirectional kutarisisa. Panguva yechizvarwa inotanga kubva pagridi yakanyatsovharwa uye inosarudza munhamba yakatarwa yekudzokorora (kazhinji 8 kusvika ku12). Nhanho yega yega inofanotaura chiratidzo chese chakavharwa, inochengeta fungidziro yekuvimbika kwepamusoro, uye inovharazve mamwe ese kutenderera kunotevera. Izvi zvinogadzira mifananidzo yemhando yepamusoro mune ingangoita kurongeka kwematanho mashoma pane autoregressive decoding.
Technical Insight
Chinhu chakakosha ndechekuvimba-kwakavakirwa masking hurongwa. Rondedzero yecosine inosarudza kuti mangani matokeni ekuburitsa imwe neimwe iteration, kutanga zvishoma uye nekumhanyisa. Nekuti kutarisa kune bidirectional, chiratidzo chega chega chinoona mufananidzo wese, saka kuita fungidziro ine chivimbo chekutanga kunotendera nhanho dzinotevera pamamiriro akasimba, sekunge kugadzirisa zvikamu zviri nyore zvepuzzle pamberi pezvisina kujeka.
Kubata MaskGIT Parallel Token Decoding
MaskGIT inogadzira mifananidzo nekufanotaura zviratidzo zvakawanda kamwechete uye kuzadza kune vane chivimbo zvakanyanya kutanga, kutsiva chizvarwa chinononoka kubva kuruboshwe kuenda-kurudyi nematanho mashoma anoenderana. MaskGIT Parallel Token Decoding ndeyekombuta-kuona mafambiro anodudzira kana kugadzira midhiya yekuona yekuongorora, mashandiro, uye kugadzira. Kuti uvake kunzwisisa kwakadzama, bata MaskGIT Parallel Token Decoding semuenzaniso wekushandisa, kwete chinhu chimwe chete: tsanangura zvinodiwa, kujekesa fungidziro, uye patsanura izvo zvingaitwe nehurongwa hwakavimbika kubva kune zvichiri kuda kutonga kwenyanzvi.
Mukuita, zvikwata zvakasimba zvinoshandisa MaskGIT Parallel Token Decoding chiyero chechokwadi nemashandiro anoita semhando yedata, kusiyana kwemwenje, uye kuenderana kwemazita. Ivo vanonyora zvakajeka maitiro ebudiriro, bvunzo vachipokana ne data rechokwadi uye mafambiro ebasa, uye iterate zvichibva pane zvakacherechedzwa maitiro ekutadza kwete kuhwina-nguva imwe chete yebhenji. Apa ndipo apo kunzwisisa kwe theoretical kunoshanduka kuve kugona kwakasimba pane chigadzirwa, mutemo, uye mashandiro.
Visual AI inogona kuita otomatiki yekuongorora, yekuona, uye yekumaka mabasa pachiyero. Panguva imwecheteyo, kodzero dzeMufananidzo uye kubvumirwa kunogona kuve njodzi dzepamutemo kana hunhu husina kujeka. Nzira yakatsiga ndeyekubatanidza kukurumidza kuyedza nekutonga: mhanyisa vatyairi vendege, tora humbowo, buritsa matanda esarudzo, uye urambe uchivandudza chengetedzo semaitiro emuenzaniso, zvinotarisirwa nemushandisi, uye zvinodikanwa zvekutonga.
Strategic Impact
Visual AI inogona kuita otomatiki yekuongorora, yekuona, uye yekumaka mabasa pachiyero.
Visual AI inogona kuita otomatiki yekuongorora, yekuona, uye yekumaka mabasa pachiyero. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Zvikwata zvekugadzira zvinogona prototype pfungwa nekukurumidza nekudzokororwa kwemaoko mashoma.
Zvikwata zvekugadzira zvinogona prototype pfungwa nekukurumidza nekudzokororwa kwemaoko mashoma. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Mashandisirwo anogona kushandisa masaini emifananidzo nemavhidhiyo ayo aimbove akaoma kugadzirisa.
Mashandisirwo anogona kushandisa masaini emifananidzo nemavhidhiyo ayo aimbove akaoma kugadzirisa. Mukutumirwa kwemhando yepamusoro, izvi zvinoshandurirwa kuita mitemo inoyerwa yekushanda, miganhu yevaridzi, uye tsika dzekudzokorora dzinodzokororwa kuitira kuti zvikwata zvikwire kuvimba pane kukwidza kusajeka.
Real-World Implementation
Kugadzira chifananidzo chizere munenge 8 kusvika 12 nhanho dzakafanana panzvimbo yemazana e autoregressive token kufanotaura.
Kupenda nzvimbo yakafukidzwa yemufananidzo nekufanotaura zvakare chete ma tokeni akavanzika ane mamiriro akatenderedza
Kirasi-yemamiriro emifananidzo synthesis paImageNet pamhando yemakwikwi ane anononoka modhi
Kushanda semusana wekudhikodha wemavara-kune-mufananidzo masisitimu akaita se Google's MUSE inoda kugadzirwa nekukurumidza.
Maitiro Ekuita
MaskGIT Parallel Token Decoding mukuita
Kugadzira mufananidzo wakazara mune 8 kusvika 12 nhanho dzakafanana panzvimbo yemazana e autoregressive token kufanotaura.
Kugadzira mufananidzo uzere munenge 8 kusvika ku12 nhanho dzakafanana pachinzvimbo chemazana ekufanotaura kwechiratidzo Matimu anowanzo kuwana mibairo iri nani kana achinge atsanangura mabindu emhando kumberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
MaskGIT Parallel Token Decoding mukuita
Kupenda nzvimbo yakafukidzwa yemufananidzo nekufanotaura zvakare chete ma tokeni akavanzika ane mamiriro akatenderedza.
Kupenda nzvimbo yakafukidzwa yemufananidzo nekufanotaura zvakare chete ma tokeni akavanzika ane mamiriro akatenderedza Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi emupendero, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
MaskGIT Parallel Token Decoding mukuita
Kirasi-yemamiriro emifananidzo synthesis paImageNet pamhando yemakwikwi ane anononoka modhi.
Kirasi-yemamiriro emifananidzo synthesis paImageNet pamhando yemakwikwi neanononoka mamodheru Matimu anowanzo kuwana mhedzisiro iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa nemitengo yekukanganisa nekufamba kwenguva.
MaskGIT Parallel Token Decoding mukuita
Inoshanda semusana wekudhikodha wemavara-kune-mufananidzo masisitimu akaita se Google's MUSE inoda kugadzirwa nekukurumidza.
Kushanda semusana wekudhirodha wemavara-kune-mufananidzo masisitimu se Google's MUSE inoda kukurumidza chizvarwa Matimu anowanzo kuwana mibairo iri nani kana achinge atsanangura emhando yepamusoro kumberi, chengetedza nzira yekukwira kwevanhu yemakesi ekumucheto, uye kuteedzera zvese zvakawanikwa zvechigadzirwa uye mutengo wekukanganisa nekufamba kwenguva.
Njodzi & Guardrails
Kodzero dzemifananidzo uye mvumo inogona kuve njodzi dzepamutemo kana provenance isina kujeka.
Kuita kwemuenzaniso kunogona kusiyanisa kupenya, huwandu hwevanhu, uye nharaunda.
Manyepo enhema anogona kusacherechedzwa kunze kwekunge zvikumbaridzo zvekuvimba zvikatariswa.
Implementation Roadmap
Tsanangura maitiro ekugamuchirwa echokwadi, kurangarira, uye mutengo wekukanganisa.
Tsanangura maitiro ekugamuchirwa echokwadi, kurangarira, uye mutengo wekukanganisa. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Edzai nedata rinoenderana nemamiriro chaiwo ekugadzira.
Edzai nedata rinoenderana nemamiriro chaiwo ekugadzira. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Wedzera ongororo yemunhu kune yakaderera-kusavimbika kana yakakwirira-inokanganisa kufanotaura.
Wedzera ongororo yemunhu kune yakaderera-kusavimbika kana yakakwirira-inokanganisa kufanotaura. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.
Tevera modhi yekudonha uye simbisa mushure mekuchinja kwekamera kana dataset.
Tevera modhi yekudonha uye simbisa mushure mekuchinja kwekamera kana dataset. Bata nhanho yega yega segedhi rehumbowo: kana maitiro asina kusangana, imbomira kuburitsa, vhara gaka, uye wobva wawedzera kushandiswa.