Ndi ntchito yofunika komanso yofunikira pakuwona pakompyuta ndi zithunzi kuti apange makanema ojambula apamwamba kwambiri.
Ngakhale mitundu ingapo yothandiza ya chithunzithunzi cha toonification yotengera mawonekedwe amphamvu a StyleGAN aperekedwa, njira zopangira zithunzizi zimakhala ndi zovuta zomveka zikagwiritsidwa ntchito ndi makanema, monga kukula kwa chimango chokhazikika, kufunikira kolumikizana ndi nkhope, kusakhalapo kwazinthu zosakhudzana ndi nkhope. , ndi kusagwirizana kwakanthawi.
Chosinthira cha VToonify chimagwiritsidwa ntchito kuthana ndi kusamutsa kwamakanema amtundu wapamwamba kwambiri.
Tiwona kafukufuku waposachedwa kwambiri pa VToonify m'nkhaniyi, kuphatikiza magwiridwe ake, zovuta zake, ndi zina.
Kodi Vtoonify ndi chiyani?
Mawonekedwe a VToonify amalola kufalitsa makanema owoneka bwino kwambiri.
VToonify imagwiritsa ntchito zigawo zapakati komanso zowoneka bwino za StyleGAN kuti ipange zithunzi zaluso zapamwamba kutengera mawonekedwe amitundu ingapo omwe amabwezedwa ndi encoder kuti asunge zambiri zamafelemu.
Zotsatira zake zomangika bwino zimatengera nkhope zosagwirizana m'makanema amitundu yosiyanasiyana monga zolowetsa, zomwe zimapangitsa kuti zigawo za nkhope zonse ziziyenda zenizeni pazotulutsa.
Chimangochi chimagwirizana ndi mitundu yaposachedwa ya StyleGAN-based image toonification, kuwalola kuti iwonjezeke ku kanema toonification, ndipo adzalandira makhalidwe okongola monga mtundu wosinthika komanso makonda ake.
izi phunziro imayambitsa zongopeka ziwiri za VToonify kutengera Toonify ndi DualStyleGAN potengera mavidiyo otengera komanso zitsanzo, motsatana.
Zoyeserera zambiri zikuwonetsa kuti dongosolo la VToonify lomwe likufunsidwa limapambana njira zomwe zilipo popanga makanema apamwamba kwambiri, ogwirizana kwakanthawi okhala ndi magawo osiyanasiyana.
Ochita kafukufuku amapereka Google Colab notebook, kotero inu mukhoza kuyipitsa manja anu pa izo.
Kodi ntchito?
Kuti mukwaniritse kusintha kosinthika kwamakanema azithunzi, VToonify imaphatikiza zabwino zamakina omasulira zithunzi ndi mawonekedwe ozikidwa pa StyleGAN.
Kuti mukhale ndi kukula kosiyanasiyana, makina omasulira zithunzi amagwiritsa ntchito maukonde osinthika. Komano, kuphunzitsa koyambira pachimake kumapangitsa kuti kufalikira kwapamwamba komanso koyendetsedwa bwino sikutheke.
Mtundu wophunzitsidwa kale wa StyleGAN umagwiritsidwa ntchito mu StyleGAN-based framework for high-resolution and controlled style transfer, ngakhale imangokhala ndi kukula kwa chithunzi chokhazikika ndi kutayika kwatsatanetsatane.
StyleGAN imasinthidwa mu hybrid framework ndikuchotsa mawonekedwe ake oyika kukula kwake ndi zigawo zotsika, zomwe zimapangitsa kuti pakhale kamangidwe kake kosinthika kofanana ndi kamangidwe kazithunzi.
Kuti musunge zambiri za chimango, phunzitsani encoder kuti achotse mawonekedwe amitundu ingapo ya chimango cholowetsamo ngati chinthu chowonjezera pa jenereta. Vtoonify amatenga kusinthasintha kwa mawonekedwe a StyleGAN poyiyika mu jenereta kuti isungunuke zonse ndi mtundu wake.
Zochepa za StyleGAN & Proposed Vtoonify
Zithunzi zaluso ndizofala m'miyoyo yathu yatsiku ndi tsiku komanso m'mabizinesi opanga zinthu monga zaluso, chikhalidwe TV ma avatar, makanema, zotsatsa zosangalatsa, ndi zina zotero.
Ndi chitukuko cha kuphunzira kwakukulu ukadaulo, tsopano ndizotheka kupanga zithunzi zaluso zapamwamba kuchokera pazithunzi zenizeni zenizeni pogwiritsa ntchito mawonekedwe osinthira azithunzi.
Pali njira zingapo zopambana zomwe zimapangidwira kutengera mawonekedwe azithunzi, ambiri omwe amapezeka mosavuta kwa ogwiritsa ntchito omwe akuyamba kugwiritsa ntchito mafoni. Makanema adakhala gawo lalikulu lazakudya zathu zapa media pazaka zingapo zapitazi.
Kuwonjezeka kwa mafilimu ochezera a pa Intaneti ndi mafilimu a ephemeral kwawonjezera kufunikira kwa kusintha kwamavidiyo mwatsopano, monga kusamutsa mavidiyo azithunzi, kuti apange mavidiyo opambana komanso osangalatsa.
Njira zomwe zilipo zokhala ndi zithunzi zimakhala ndi zovuta zazikulu zikagwiritsidwa ntchito pamakanema, zomwe zimalepheretsa kufunikira kwake pamakongoletsedwe amavidiyo azithunzi.
StyleGAN ndi msana wamba popanga mawonekedwe osinthira zithunzi zazithunzi chifukwa cha kuthekera kwake kupanga nkhope zapamwamba zowongolera masitayilo osinthika.
Dongosolo lochokera ku StyleGAN (lomwe limadziwikanso kuti chithunzi toonification) limayika nkhope yeniyeni mu malo obisika a StyleGAN kenako ndikuyika kachidindo kamene kamatsatira ku StyleGAN ina yokonzedwa bwino pazithunzi zajambula kuti apange mawonekedwe ojambulidwa.
StyleGAN imapanga zithunzi zokhala ndi nkhope zofananira komanso kukula kokhazikika, zomwe sizimakonda mawonekedwe amtundu weniweni. Kudulira nkhope ndi kuyanjanitsa muvidiyoyi nthawi zina kumapangitsa kuti pakhale nkhope yapang'onopang'ono komanso mawonekedwe osasangalatsa. Ochita kafukufuku amatcha nkhaniyi StyleGAN's 'fixed-crop restriction.'
Kwa nkhope zosasinthika, StyleGAN3 yaperekedwa; komabe, imangogwirizira kukula kwa chithunzi.
Kuphatikiza apo, kafukufuku waposachedwa adapeza kuti kusungitsa nkhope zosagwirizana ndizovuta kwambiri kuposa nkhope zolumikizana. Kuyika nkhope kolakwika kumawononga kusamutsa mawonekedwe, zomwe zimapangitsa kuti pakhale zovuta monga kusintha dzina ndi zina zomwe zikusowa pamafelemu omangidwanso.
Monga momwe tafotokozera, njira yabwino yosinthira mavidiyo azithunzi iyenera kuthana ndi izi:
- Kuti musunge mayendedwe enieni, njirayo iyenera kuthana ndi nkhope zosasinthika komanso mavidiyo osiyanasiyana. Kanema wamkulu, kapena mawonekedwe akulu, amatha kujambula zambiri ndikusunga nkhope kuti isasunthike.
- Kuti mupikisane ndi zida zamakono za HD zomwe zimagwiritsidwa ntchito kwambiri masiku ano, kanema wapamwamba kwambiri ndiyofunikira.
- Kuwongolera masitayelo osinthika kuyenera kuperekedwa kuti ogwiritsa ntchito asinthe ndikusankha zomwe akufuna popanga njira yolumikizirana ndi ogwiritsa ntchito.
Kuti izi zitheke, ofufuza akuwonetsa VToonify, njira yosakanizidwa yosakanizidwa yamavidiyo. Kuti athe kuthana ndi vuto lokhazikika la mbewu, ofufuza amayamba kuphunzira kumasulira kofanana mu StyleGAN.
VToonify imaphatikiza zabwino zamamangidwe ozikidwa pa StyleGAN ndi mawonekedwe omasulira zithunzi kuti akwaniritse kusintha kosinthika kwamakanema azithunzi.
Zotsatirazi ndizothandiza kwambiri:
- Ofufuza amafufuza zazovuta za StyleGAN za mbewu zokhazikika ndikupereka yankho potengera kufanana kwa zomasulira.
- Ofufuzawo amapereka mawonekedwe apadera a VToonify osinthika amakanema amakanema apamwamba kwambiri omwe amathandizira nkhope zosalumikizana komanso mavidiyo osiyanasiyana.
- Ochita kafukufuku amapanga VToonify pazitseko za Toonify ndi DualStyleGAN ndikugwirizanitsa misana malinga ndi deta ndi chitsanzo kuti athe kusamutsa mavidiyo otengera kusonkhanitsa ndi zitsanzo.
Poyerekeza Vtoonify ndi zitsanzo zina zamakono
Toonify
Imakhala ngati maziko osinthira masitayelo otengera kusonkhanitsa pankhope zolumikizana pogwiritsa ntchito StyleGAN. Kuti atengenso masitayelo, ofufuza akuyenera kugwirizanitsa nkhope ndikudula zithunzi 256256 za PSP. Toonify imagwiritsidwa ntchito kupanga zotsatira zokongoletsedwa ndi ma code 1024 * 1024.
Pomaliza, amayanjanitsanso zotsatira za kanemayo kumalo ake enieni. Malo osasinthidwa asinthidwa kukhala akuda.
DualStyleGAN
Ndi msana wa kusamutsa kotengera zitsanzo kutengera StyleGAN. Amagwiritsa ntchito njira zomwezo zisanachitike komanso pambuyo pokonza monga Toonify.
Zithunzi za Pix2pixHD
Ndi mtundu womasulira kuchokera kuzithunzi kupita ku chithunzi womwe umagwiritsidwa ntchito kwambiri kufupikitsa mamotchi ophunzitsidwa kale kuti asinthe mawonekedwe apamwamba. Amaphunzitsidwa pogwiritsa ntchito ma data awiriawiri.
Ofufuza amagwiritsa ntchito pix2pixHD monga zowonjezera mamapu ake chifukwa amagwiritsa ntchito mapu ojambulidwa.
Choyamba Order Motion
FOM ndi mtundu wofananira wa makanema ojambula. Idaphunzitsidwa pazithunzi za 256256 ndipo imachita bwino ndi makulidwe ena azithunzi. Zotsatira zake, ofufuza amayamba amakweza mafelemu amakanema kukhala 256*256 kuti FOM akhale makanema ojambula kenako ndikusinthanso kukula kwake koyambirira.
Kuyerekeza koyenera, FOM imagwiritsa ntchito mawonekedwe ake oyambira ngati mawonekedwe ake.
DaGAN
Ndi 3D face animation model. Amagwiritsa ntchito njira zomwezo zokonzekera deta ndi postprocessing monga FOM.
ubwino
- Itha kugwiritsidwa ntchito mu zaluso, ma avatar azama TV, makanema, kutsatsa zosangalatsa, ndi zina zotero.
- Vtoonify itha kugwiritsidwanso ntchito mu metaverse.
sitingathe
- Njirayi imachotsa zonse zomwe zili ndi deta ndi chitsanzo kuchokera kuzitsulo zamtundu wa StyleGAN, zomwe zimapangitsa kuti pakhale tsankho lachitsanzo.
- Zopangidwazo zimachitika makamaka chifukwa cha kusiyana kwa kukula pakati pa chigawo cha nkhope chokongoletsedwa ndi zigawo zina.
- Njirayi imakhala yocheperapo pochita ndi zinthu zomwe zili m'dera la nkhope.
Kutsiliza
Pomaliza, VToonify ndi chimango chowongolera makanema owongolera kwambiri.
Dongosololi limagwira ntchito bwino pakuwongolera makanema ndipo limathandizira kuwongolera bwino mawonekedwe, mawonekedwe amtundu, ndi digiri ya masitayilo pochepetsa mitundu yazithunzi zozikidwa pa StyleGAN molingana ndi onse awo. zopangapanga ndi ma network.
Siyani Mumakonda