{"id":105302,"date":"2022-10-13T15:36:48","date_gmt":"2022-10-13T13:36:48","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/otkryt-kod-sistemy-mashinnogo-obucheniya-dlya-generaczii-realistichnyh-dvizhenij-cheloveka"},"modified":"2022-10-13T15:36:48","modified_gmt":"2022-10-13T13:36:48","slug":"otkryt-kod-sistemy-mashinnogo-obucheniya-dlya-generaczii-realistichnyh-dvizhenij-cheloveka","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/news\/otkryt-kod-sistemy-mashinnogo-obucheniya-dlya-generaczii-realistichnyh-dvizhenij-cheloveka","title":{"rendered":"\u00cbsht\u00eb hapur kodi i sistemit t\u00eb m\u00ebsimit automatik p\u00ebr gjenerimin e l\u00ebvizjeve realiste t\u00eb njeriut.","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Nj\u00eb grup hulumtuesish nga Universiteti i Tel Avivit zbuloi tekstet origjinale t\u00eb lidhura me sistemin e m\u00ebsimit t\u00eb makinave MDM (Motion Diffusion Model), i cili lejon t\u00eb gjenerohen l\u00ebvizje realiste t\u00eb njeriut. Kodi \u00ebsht\u00eb i shkruar n\u00eb gjuh\u00ebn Python duke p\u00ebrdorur korniz\u00ebn PyTorch dhe shp\u00ebrndahet n\u00ebn licenc\u00ebn MIT. P\u00ebr t\u00eb kryer eksperimente mund t\u00eb p\u00ebrdoren si modelet e gatshme, ashtu edhe t\u00eb realizohet trajnimi i modeleve nga vet\u00eb, duke p\u00ebrdorur skriptet e ofruara, p\u00ebr shembull, duke p\u00ebrdorur koleksionin e imazheve 3D t\u00eb njeriut HumanML3D. Nj\u00eb GPU me mb\u00ebshtetje p\u00ebr CUDA \u00ebsht\u00eb e nevojshme p\u00ebr trajnimin e sistemit. <\/p>\n
P\u00ebrdorimi i mund\u00ebsive tradicionale p\u00ebr t\u00eb animuar l\u00ebvizjet e njeriut \u00ebsht\u00eb i v\u00ebshtir\u00eb p\u00ebr shkak t\u00eb nd\u00ebrlikimeve q\u00eb lidhen me diversitetin e madh t\u00eb l\u00ebvizjeve t\u00eb mundshme dhe v\u00ebshtir\u00ebsin\u00eb e p\u00ebrshkrimit t\u00eb tyre formal, si dhe p\u00ebr shkak t\u00eb ndjeshm\u00ebris\u00eb s\u00eb madhe t\u00eb perceptimit t\u00eb njeriut ndaj l\u00ebvizjeve jo natyrale. P\u00ebrpjekjet e m\u00ebparshme p\u00ebr t\u00eb p\u00ebrdorur modelet gjeneruese t\u00eb m\u00ebsimit t\u00eb makinave kishin probleme me cil\u00ebsin\u00eb dhe shprehshm\u00ebrin\u00eb e kufizuar. <\/p>\n
N\u00eb sistemin e propozuar \u00ebsht\u00eb b\u00ebr\u00eb nj\u00eb p\u00ebrpjekje p\u00ebr t\u00eb p\u00ebrdorur modelet e difuzionit p\u00ebr t\u00eb gjeneruar l\u00ebvizje, t\u00eb cilat n\u00eb thelb jan\u00eb m\u00eb t\u00eb p\u00ebrshtatshme p\u00ebr simullimin e l\u00ebvizjeve njer\u00ebzore, por nuk jan\u00eb pa mang\u00ebsi, si\u00e7 jan\u00eb k\u00ebrkesat e larta p\u00ebr burime kompjuterike dhe complexiteti menaxhimit. P\u00ebr t\u00eb minimizuar mang\u00ebsit\u00eb e modeleve t\u00eb difuzionit, n\u00eb MDM \u00ebsht\u00eb angazhuar nj\u00eb rrjet nervor me arkitektur\u00ebn \u00abtransformer\u00bb dhe parashikimi i nj\u00eb mostre n\u00eb vend t\u00eb parashikimit t\u00eb zhurm\u00ebs n\u00eb \u00e7do hap, gj\u00eb q\u00eb leht\u00ebson parandalimin e anomali-ve, si humbja e kontaktit t\u00eb sip\u00ebrfaqes me k\u00ebmb\u00ebn. <\/p>\n
P\u00ebr menaxhimin e gjenerimit \u00ebsht\u00eb parashikuar mund\u00ebsia e aplikimit t\u00eb nj\u00eb p\u00ebrshkrimi tekstual t\u00eb veprimit n\u00eb nj\u00eb gjuh\u00eb natyrale (p.sh., \u00abnj\u00eb njeri po ec\u00ebn p\u00ebrpara dhe p\u00ebrkul\u00ebt p\u00ebr t\u00eb ngritur di\u00e7ka nga toka\u00bb) ose p\u00ebrdorimi i veprimeve tipike, si \u00abvrapimi\u00bb dhe \u00abk\u00ebrcimi\u00bb. Sistemi mund t\u00eb p\u00ebrdoret gjithashtu p\u00ebr redaktimin e l\u00ebvizjeve dhe plot\u00ebsimin e detajeve t\u00eb humbura. Studiuesit kryen teste, ku pjes\u00ebmarr\u00ebsve iu k\u00ebrkua t\u00eb zgjidhnin rezultatet m\u00eb cil\u00ebsore nga disa opsione \u2014 n\u00eb 42% t\u00eb rasteve, njer\u00ebzit preferuan l\u00ebvizjet e sintetizuara sesa ato reale.
![\"Luaj](\"https:\/\/img.youtube.com\/vi\/rVkIDj5wgjs\/hqdefault.jpg\")
\r\n