Uhlolojikelele
I-DDPM ne-DDIM izindlela ezimbili zokuqalisa inqubo ehlehlayo yemodeli yokusabalalisa, ukuguqula umsindo ongahleliwe wenze isithombe ngesinyathelo ngesinyathelo. I-DDPM iresiphi yokuqala ye-stochastic; I-DDIM iyisinqamuleli esisheshayo, esinqumayo esikhiqiza izithombe ezifanayo ngezinyathelo ezimbalwa kakhulu.
I-DDPM ne-DDIM Samplers ingeyokugeleza komsebenzi wokubona ngekhompyutha okuhumusha noma okukhiqiza imidiya ebonakalayo ukuze ihlaziywe, isebenze, futhi isungule.
I-Deep Dive
Imodeli yokusabalalisa iqeqeshwa ngokwengeza kancane kancane umsindo we-Gaussian ezithombeni, bese ifunda ukubikezela lowo msindo. Ukuthatha amasampula kubuyisela emuva lokhu. I-DDPM (Denoising Diffusion Probabilistic Models, Ho et al. 2020) ihamba ibuyela emuva kuwo wonke amaleveli omsindo, yengeza i-dab entsha yomsindo ongahleliwe esinyathelweni ngasinye, ngakho-ke ivamise ukudinga izinyathelo ezingamakhulu ukuya enkulungwaneni. I-DDIM (I-Denoising Diffusion Implicit Models, Ingoma et al. 2021) isebenzisa kabusha inethiwekhi efanayo ncamashi eqeqeshiwe kodwa ilandela i-non-Markovian, i-deterministic trajectory. Ngokuyeka ukungahleliwe okujovwe, i-DDIM ingakwazi ukweqa izikhathi eziningi futhi isahlala esithombeni sekhwalithi ephezulu ngezinyathelo ezingu-10-50. Ngenxa yokuthi i-DDIM iyanquma, umsindo ofanayo wokuqala uhlala uveza isithombe esifanayo, uvumela ukuhumusha okushelelayo nokukhiqiza kabusha.
I-Technical Insight
Womabili amasampula asebenzisa inethiwekhi ebikezela umsindo we-epsilon owengezwe esithombeni ngesikhathi sika-t. Isibuyekezo se-DDPM sikhipha inguqulo enezikali yaleso sibikezelo bese sengeza umsindo ohlukile othathwe ngemuva. I-DDIM ibhala kabusha isibuyekezo ukuze iqale ilinganisele isithombe esihlanzekile esingu-x0, bese isiphinda isiqhubekisele phambili esinyathelweni sesikhathi esilandelayo (esincane) ngaphandle kwetemu le-stochastic. Ipharamitha eta ihlanganisa kokubili: eta=1 ithola i-DDPM, i-eta=0 inikeza i-DDIM enquma ngokugcwele.
Ukwazi kahle amasampula e-DDPM kanye ne-DDIM
I-DDPM ne-DDIM izindlela ezimbili zokuqalisa inqubo ehlehlayo yemodeli yokusabalalisa, ukuguqula umsindo ongahleliwe wenze isithombe ngesinyathelo ngesinyathelo. I-DDPM iresiphi yokuqala ye-stochastic; I-DDIM iyisinqamuleli esisheshayo, esinqumayo esikhiqiza izithombe ezifanayo ngezinyathelo ezimbalwa kakhulu. I-DDPM ne-DDIM Samplers ingeyokugeleza komsebenzi wokubona ngekhompyutha okuhumusha noma okukhiqiza imidiya ebonakalayo ukuze ihlaziywe, isebenze, futhi isungule. Ukuze wakhe ukuqonda okujulile, phatha i-DDPM ne-DDIM Samplers njengemodeli yokusebenza, hhayi isici esisodwa: chaza imiphumela efiselekayo, ucacise ukucabanga, futhi uhlukanise lokho isistimu engakwenza ngokwethembeka kulokho okusadinga ukwahlulela kochwepheshe.
Empeleni, amaqembu aqinile asebenzisa i-DDPM ne-DDIM Samplers ukunemba kwebhalansi namaqiniso okusebenza njengekhwalithi yedatha, ukuhluka kokukhanya, nokuvumelana kwamalebula. Babhala imibandela yempumelelo ecacile, ukuhlola okuqhathaniswa nedatha engokoqobo nokugeleza komsebenzi, futhi baphindaphinde ngokusekelwe kumaphethini okuhluleka aqashiwe esikhundleni sokuwina kwebhentshimakhi yesikhathi esisodwa. Yilapho ukuqonda kwethiyori kuguquka kube amandla ahlala njalo kuwo wonke umkhiqizo, inqubomgomo, kanye nokusebenza.
I-Visual AI ingakwazi ukuhlola, ukutholwa, nokumaka imisebenzi esikalini. Ngesikhathi esifanayo, amalungelo ezithombe kanye nemvume kungaba ubungozi bomthetho uma ukutholakala kungacacile. Indlela eqine kakhulu iwukuhlanganisa isivinini sokuhlola nesiyalo sokuphatha: qhuba abashayeli bezindiza, bamba ubufakazi, ushicilele amalogi ezinqumo, futhi ubuyekeze izivikelo ngokuqhubekayo njengoba imodeli yokuziphatha, okulindelwe ngabasebenzisi, kanye nezimfuneko zokulawula zishintsha.
I-Strategic Impact
I-Visual AI ingakwazi ukuhlola, ukutholwa, nokumaka imisebenzi esikalini.
I-Visual AI ingakwazi ukuhlola, ukutholwa, nokumaka imisebenzi esikalini. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.
Amathimba aqanjiwe angakwazi ukulinganisa imiqondo ngokushesha ngezibuyekezo ezimbalwa ezenziwa mathupha.
Amathimba aqanjiwe angakwazi ukulinganisa imiqondo ngokushesha ngezibuyekezo ezimbalwa ezenziwa mathupha. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.
Imisebenzi ingasebenzisa amasiginali wesithombe nawevidiyo obekunzima ukuwenza ngaphambilini.
Imisebenzi ingasebenzisa amasiginali wesithombe nawevidiyo obekunzima ukuwenza ngaphambilini. Ekusetshenzisweni kwekhwalithi ephezulu, lokhu kuhunyushwa emithethweni yokusebenza elinganisekayo, imingcele yobunikazi, nemikhuba yokubuyekeza ephindelelayo ukuze amaqembu akwazi ukukala ukuzethemba esikhundleni sokukala ukungaqondakali.
Ukuqaliswa Komhlaba Wangempela
Ukwenziwa kwesithombe esine-Stable Diffusion, lapho i-DDIM inikezwa khona njengesampula ezenzakalelayo esheshayo yokwaziswa kombhalo kuya kwesithombe kumathuluzi afana ne-Automatic1111 ne-ComfyUI.
Amapayipi obuciko aphindaphindekayo alungisa imbewu engahleliwe nge-deterministic DDIM ukuze umyalo ofanayo kanye nembewu ihlale ikhiqiza isithombe esifanayo.
Ukuhumusha okushelelayo kwesikhala esifihlekile phakathi kwezithombe ezimbili zokugqwayiza kwe-morphing, okwenziwe kwaba nokwenzeka yimephu yokunquma ye-DDIM ukusuka kumsindo kuye kokuphumayo.
Ukuphindwaphindwa kokudala okusheshayo lapho abaklami basebenzisa ukuhlola kuqala kwe-DDIM okuyizinyathelo ezingu-20 ukuze bahlole imiqondo ngaphambi kokuzibophezela ekunikezeni okunensayo, okuthembekile okuphezulu okugcwele.
Amaphethini Okusebenzisa
Amasampula e-DDPM kanye ne-DDIM ayasebenza
Ukwenziwa kwesithombe esine-Stable Diffusion, lapho i-DDIM inikezwa khona njengesampula ezenzakalelayo esheshayo yokwaziswa kombhalo kuya kwesithombe kumathuluzi afana ne-Automatic1111 ne-ComfyUI.
Ukukhiqizwa kwesithombe esine-Stable Diffusion, lapho i-DDIM inikezwa khona njengesampulu esizenzakalelayo esisheshayo sokwaziswa kombhalo kuya kwesithombe kumathuluzi afana ne-Automatic1111 kanye namaQembu e-ComfyUI ngokuvamile athola imiphumela engcono uma echaza imingcele yekhwalithi ngaphambili, agcine indlela yokukhuphuka komuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.
Amasampula e-DDPM kanye ne-DDIM ayasebenza
Amapayipi obuciko aphindaphindekayo alungisa imbewu engahleliwe nge-deterministic DDIM ukuze umyalo ofanayo kanye nembewu ihlale ikhiqiza isithombe esifanayo.
Amapayipi obuciko aphindaphindekayo alungisa imbewu engahleliwe nge-DDIM enqumayo ukuze ngokushesha okufanayo kanye nembewu ihlale ikhiqiza isithombe esifanayo Amaqembu ngokuvamile athola imiphumela engcono uma echaza imikhawulo yekhwalithi ngaphambili, egcina indlela yokukhuphuka yabantu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.
Amasampula e-DDPM kanye ne-DDIM ayasebenza
Ukuhumusha okushelelayo kwesikhala esifihlekile phakathi kwezithombe ezimbili zokugqwayiza kwe-morphing, okwenziwe kwaba nokwenzeka yimephu yokunquma ye-DDIM ukusuka kumsindo kuye kokuphumayo.
Ukuhunyushwa okushelelayo kwesikhala esifihlekile phakathi kwezithombe ezimbili zokugqwayiza kwe-morphing, okwenziwe kwaba nokwenzeka ukunqunywa kwemephu ye-DDIM kusuka emsindweni kuya kokuphumayo Amaqembu ngokuvamile athola imiphumela engcono uma echaza imingcele yekhwalithi ngaphambili, agcina indlela yokukhuphuka komuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.
Amasampula e-DDPM kanye ne-DDIM ayasebenza
Ukuphindwaphindwa kokudala okusheshayo lapho abaklami basebenzisa ukuhlola kuqala kwe-DDIM okuyizinyathelo ezingu-20 ukuze bahlole imiqondo ngaphambi kokuzibophezela ekunikezeni okunensayo, okuthembekile okuphezulu okugcwele.
Ukuphindwaphindwa okushesha kokuqamba lapho abaklami basebenzisa ukuhlola kuqala kwe-DDIM okuzinyathelo ezingu-20 ukuze bahlole imiqondo ngaphambi kokuzinikela ekunikezeni isinyathelo esigcwele esinensa, nesithembeke kakhulu Amathimba ngokuvamile athola imiphumela engcono uma echaza imingcele yekhwalithi ngaphambili, agcina indlela yokukhuphuka komuntu yamakesi asemaphethelweni, futhi alandelele kokubili izinzuzo zokukhiqiza nezindleko zamaphutha ngokuhamba kwesikhathi.
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.
Chaza indlela yokwamukela yokunemba, ukukhumbula, nezindleko zamaphutha. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.
Hlola ngedatha efana nezimo zangempela zokukhiqiza.
Hlola ngedatha efana nezimo zangempela zokukhiqiza. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.
Engeza isibuyekezo somuntu ukuze uthole ukuzethemba okuphansi noma izibikezelo zomthelela omkhulu.
Engeza isibuyekezo somuntu ukuze uthole ukuzethemba okuphansi noma izibikezelo zomthelela omkhulu. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.
Landelela ukukhukhuleka kwemodeli bese uqinisekisa kabusha ngemva kwezinguquko zekhamera noma zesethi yedatha.
Landelela ukukhukhuleka kwemodeli bese uqinisekisa kabusha ngemva kwezinguquko zekhamera noma zesethi yedatha. Phatha isinyathelo ngasinye njengesango lobufakazi: uma imibandela ingafinyelelwa, misa ukukhishwa, vala igebe, bese unweba ukusetshenziswa.