InoteveraGaidhi rinotevera
Maitiro ekugadzirisa Vhidhiyo neKugadzirisa Chinyorwa neAI
AI Yekutarisa
Gwara reVisual AI
Text-to-motion AI inogadzira 3D character animation kubva pakukurumidza kunyorwa sekuti "munhu anosvetuka pamusoro pemvura," achigadzira nhevedzano yemaskeletal poses kwete vhidhiyo pixels.
Teerera kugwaro iriAI-yakarondedzerwa · ~3 min
Zvishandiso zvinoenderana zvinobvisa kufambisa kutorwa kubva kune yakajairwa foni vhidhiyo. Pamwe chete vanopa vanogadzira indie, vanosimudzira uye vadzidzisi kukurumidza kwekutanga dhizaini yekufamba kwevatambi vasina mocap sutu kana ruoko-kiyi yenguva.
Mameseji-ku-kufamba masisitimu anogadzira miviri inomisikidza, kwete pixels. Imwe neimwe furemu inotsanangura skeleton: midzi chinzvimbo uye kutaridzika pamwe nekutenderera kwega rega rega, rinowanzo kuratidzwa pane parametric body modhi se SMPL. Klipu saka igidhi remafuremu nezvimiro, ndosaka matekiniki ekugadzira akavakirwa mifananidzo nemutauro achigona kuchinjirwa kufamba. Nzira inonyanya kuzivikanwa ndeyeMotion Diffusion Model (MDM), yakabudiswa nevatsvakurudzi paYunivhesiti yeTel Aviv muna 2022. Inotanga kubva kune ruzha rusina kurongeka rwakaumbwa sechikwata chekufamba uye rinoshandisa transformer kuti ibvise iyo ruzha nhanho nhanho, inotungamirirwa nekunyora kwemavara kubva kuCLIP. Mamodheru akaita seaya anodzidza kubva kumaseti epa data anotorwa netsanangudzo dzakanyorwa, kunyanya HumanML3D, ine zvimedu zvakatorwa zvakanyanya kubva kuAMASS kufamba kwekutora kuunganidzwa. Mhuri yechipiri inobata kufamba semutauro: T2M-GPT uye MoMask compress kufamba mune discrete tokens neVQ-VAE-maitiro quantization, wobva wagadzira iwo ma tokeni nenzira iyo modhi yemutauro inoburitsa mazwi. Bazi rakasiyana rinosvetuka zvinyorwa zvachose. AI inofamba yekutora maturusi akadai seMove.ai, DeepMotion, Rokoko Vision uye Wonder Studio (ikozvino Autodesk Flow Studio) inofungidzira 3D inomira kubva kuvhidhiyowo zvayo. Ivo vanoona makiyi emuviri e2D mune yega furemu, voasimudza mu3D, uye anotsvedzerera mhedzisiro nekufamba kwenguva. Kune akawanda mapurojekiti izvi zvinotsiva masutu ekutora uye mamaki masitudiyo. Muzviitiko zvese zviri zviviri zvinobuda zvinofanirwa kudzoserwa: kumepuwa pane hunhu hune mazita emapfupa, chiyero uye kuzorora kunosiyana kubva kune kwakabva skeleton. Imwe pfungwa isiriyo ndeyekuti maturusi aya anounza maitirwo apera. Zvimwe zvinokanganisa zvinosanganisira kutsvedza kwetsoka, uko tsoka dzinotsvedza pasi, tsoka dzichinyura nepasi, jitter, uye maoko akaomarara, nekuti akawanda dataset anorekodha zvigunwe zvisina kunaka. Zvikamu zvinogadzirwa zvinongogara masekonzi mashoma, uye zvinokurudzira zvinosanganisira kutevedzana kwechiito kana zvinhu zvinoramba zvakaoma. Mazhinji masitudiyo anobata zvinobuda seyekutanga kupasa iyo animator inonatsa mumidziyo yakaita seMaya, Blender kana MotionBuilder.
Mugadziri wemutambo we indie anosimudzira modhi yekufambisa ye "musoja aneta anokamhina mberi ogara pakireti," anotumira clip seFBX, uye anoidzosera kuUnity character sechibatiso pamberi pekupedzisira animation.
Idiki animation studio mafirimu mutambi muimba yekutandarira nefoni imwe, inomhanyisa mavhidhiyo kuburikidza nevhidhiyo-based motion capture sevhisi, uye inosuka marangwanda akabuda muBlender pachinzvimbo chekuhaya mamaki-based capture stage.
Muimbi wekufungidzira anovharira chiitiko chekurwa nekugadzira gumi nembiri mapfupi mameseji-kune-kufamba zvimedu, achizvibatanidza nepakati-, uye kuratidza director nguva uye nhanho isati yadhirowa chero makiyi.
Muongorori wekurapa kwemuviri anofananidza mafambiro emurwere, akatorwa kubva kukiriniki vhidhiyo ine mucherechedzo wepose estimation, achipesana nereferensi yekufamba-famba kutevedzera shanduko mumajoini makona kwemavhiki.
Tsvagiridzo iri kuenda kune zvimedu zvirefu, mavara akawanda anodyidzana, kufamba kunoremekedza zvipingaidzo zvemuviri, uye mavara anodyidzana nezvinhu uye zviratidziro pane kufamba munzvimbo isina chinhu. Injini dzemitambo uye mapakeji emupepeti ari kutanga kubatanidza chizvarwa uye kutorwa kwevhidhiyo seakavakirwa-mukati maficha, ayo anogona kuita kuti hutsinye hupopi huwanikwe kune vakawanda vanogadzira. Miganho yakakosha inoramba iripo: data yekudzidziswa idiki kana ichienzaniswa nemaseti emifananidzo, inomiririra mhando dzakawanda dzemuviri, tsika nemaitiro ekufambisa, uye kunaka kwemaoko nezviso zvinosara kumashure kwekufamba-muviri. Tarisira maanimators kuchengetedza basa repakati mukutungamira, kugadzirisa uye kuwedzera mashandiro kune akagadzirwa kufamba.
Tsamba yemahara
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Text-to-motion AI inogadzira 3D character animation kubva pakukurumidza kunyorwa sekuti "munhu anosvetuka pamusoro pemvura," achigadzira nhevedzano yemaskeletal poses kwete vhidhiyo pixels. Zvishandiso zvinoenderana zvinobvisa kufambisa kutorwa kubva kune yakajairwa foni vhidhiyo. Pamwe chete vanopa vanogadzira indie, vanosimudzira uye vadzidzisi kukurumidza kwekutanga dhizaini yekufamba kwevatambi vasina mocap sutu kana ruoko-kiyi yenguva.
Mameseji-kune-anofambisa modhi inobuda muviri inomira furemu nefuremu, kazhinji nzvimbo yemidzi uye yekumisikidza pamwe nekutenderera kwemajoini. Kupa uye dhizaini yehunhu inoitika kune imwe nzvimbo mupombi.
MDM inogadzirisa dhinoiser yayo pane chinyorwa chinomisikidzwa kubva kuCLIP, iyo inobatanidza mutauro kune zvinoreva inodzidzwa kubva pamufananidzo uye mavara maviri.
HumanML3D paviri tsananguro dzakanyorwa dzine zvimedu zvinofamba zvakadhonzwa zvakanyanya kubva kuAMASS, muunganidzwa wakakura we data yakamisikidzwa yekutora data.
Nekuda kwekuti network inoburitsa zvinokonzeresa, kudzidziswa kunogona kuranga zvakananga mhedzisiro yemuviri, senge tsoka dzinofamba dzichifanira kudyarwa.
Aya mamodheru anodzidza mazwi ezviratidzo zvinofamba uye obva agadzira kutevedzana kwezviratidzo, senge modhi yemutauro inogadzira zvinyorwa.
Ramba uchidzidza
Mamwe madhairekitori akasarudzirwa nyaya iyi
InoteveraGaidhi rinotevera
Maitiro ekugadzirisa Vhidhiyo neKugadzirisa Chinyorwa neAI
AI Yekutarisa