Ukulinganisa Ukujula Kwe-Marigold Diffusion
U-Marigold uphinda asebenzise imodeli yokusabalalisa yesithombe esiqeqeshwe kusengaphambili (I-Stable Diffusion) ukuze abikezele amamephu ajulile anemininingwane eminingi.
Uhlolojikelele
It shows you can turn a generator's rich visual knowledge into a precise perception tool with surprisingly little training data.
I-Deep Dive
U-Marigold (ETH Zurich, CVPR 2024 Best Paper Honourable Mention) uhlela kabusha ukulinganisa kokujula njengenkinga yesizukulwane esinemibandela. Esikhundleni sokuqeqesha inethiwekhi ejulile kusukela ekuqaleni, ishuna kahle i-Stable Diffusion ukuze 'ikhiqize' imephu ejulile ebekwe esithombeni sokufaka. Ukuqonda ukuthi imodeli eqeqeshelwe ukuhlanganisa izithombe ze-photorealistic isivele ifundile ijometri yesigcawu, ukukhanya, kanye nesakhiwo ekujuleni kwendawo yayo ecashile, okuyizinto ezihamba phambili eziwusizo ekujuleni. Ngokumangazayo, i-Marigold yayicutshungulwe kahle kumadathasethi okwenziwa kuphela (njenge-Hypersim kanye ne-Virtual KITTI) kodwa ihlanganisa kahle izithombe zangempela ezingasho lutho. Ikhiqiza ukujula kwesihlobo okungaguquki kwe-affine ngemininingwane emihle kakhulu, nakuba i-denoising ephindaphindayo iyenza ihambe kancane kunamamodeli okuphakela phambili njenge-DepthAnything.
I-Technical Insight
I-Marigold isebenza endaweni ecashile ye-Stable Diffusion. Kokubili isithombe nokujula kwemephu kubhalwe ngekhodi i-VAE efanayo; i-U-Net icushwe kahle ukuze isho umsindo wokujula okucashile okubekwe esithombeni esihlanzekile esicashile. Lapho kucatshangwa khona isebenzisa iluphu evamile ephindaphindayo ephindaphindayo, bese ikhipha amakhodi ukujula okucashile. Ngenxa yokuthi isampula, ukugijima okuningi kungahlanganiswa ukuze kube nokuzinza, ukuhweba kubalwe ukuze kube nokunemba. Kamuva 'i-LCM' nezinguqulo ze-distilled zesinyathelo esisodwa zisika inqwaba yezinyathelo zehle ziye endaweni eyodwa.
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 Lesilinganiso Sokujula Kwe-Marigold Diffusion
Iresiphi ye-Marigold, izimiso zokusabalalisa zokuhlelwa kahle kokubikezela okuminyene, ijwayele ukudlula ukujula kuya kokujwayelekile kwendawo, ukubola kwesithombe sangaphakathi, kanye nokulinganisa kwezinto ezibonakalayo. Okuhlukile okushesha kakhulu kwe-distilled kanye nokufana kwemodeli kuvala igebe lejubane ngamanethiwekhi okudlulisela phambili, okwenza umbono osuselwe ekusakazweni usebenze kumathuluzi asebenzisanayo. Lindela ithrendi ebanzi lapho umgogodla owodwa okhiqizayo oqeqeshelwe kusengaphambili ujwayelana nemisebenzi eminingi yejiyomethri neyokubona, unciphisa isidingo samasethi edatha anelebula anomsebenzi othile.
Ukuqaliswa Komhlaba Wangempela
Ikhipha ukujula okucolisekile ezithombeni zezakhiwo nezomkhiqizo ukuze kukhanye kabusha nemifanekiso ye-3D.
Ikhiqiza amamephu anemininingwane ephezulu asetshenziswa njengesimo sokwenziwa kwesithombe esilawulekayo nevidiyo.
Ifilimu yokusiza kanye namaqembu e-VFX emsebenzini we-matte ne-parallax lapho ukunemba komphetho kubaluleke khona.
Isebenza njengesisekelo socwaningo esibonisa ukuthi ungajwayelanisa kanjani okuphambili kokuzalayo nemisebenzi yokuqagela eminyene.
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 Marigold Diffusion Depth Estimation 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
Ukulinganisa Ukujula Kwe-Monocular
Imibuzo evame ukubuzwa
What is Marigold Diffusion Depth Estimation?
U-Marigold uphinda asebenzise imodeli yokusabalalisa yesithombe esiqeqeshwe kusengaphambili (I-Stable Diffusion) ukuze abikezele amamephu ajulile anemininingwane eminingi. Kubonisa ukuthi ungakwazi ukushintsha ulwazi olucebile olubonakalayo lwejeneretha lube ithuluzi lokubona elinembile elinedatha yokuqeqeshwa encane ngokumangalisayo.
Iyiphi imodeli eqeqeshwe kusengaphambili uMarigold yakhela phezu kwayo?
I-Marigold ishuna kahle imodeli ye-Stable Diffusion efihlekile yokusabalalisa ukuze ikhiqize amamephu ajulile.
I-Marigold iwuhlaka kanjani umsebenzi wokulinganisa ukujula?
Iphatha ukujula njengokuthile okumele 'kwenziwe' kufakwe esimweni sesithombe sokufaka, kusetshenziswa kabusha imishini yokusabalalisa.
Yini eyayimangaza ngedatha yokuqeqeshwa kukaMarigold?
I-Marigold icushwe kahle kudatha yokwenziwa efana ne-Hypersim ne-Virtual KITTI kodwa ihlanganisa i-zero-shot ezithombeni zangempela.
Kungani i-Marigold ngokuvamile ihamba kancane kunamamodeli okujula okuya phambili?
Amamodeli okusabalalisa enza umsindo ngezinyathelo eziningi, okwenza ukucatshangelwa kuhamba kancane kunephasi eyodwa eya phambili.
I-Marigold yenza kusiphi isikhala inqubo yayo yokusabalalisa?
Kokubili isithombe nokujula kubhalwe ngekhodi endaweni ecashile ye-VAE lapho i-U-Net izwakala khona.