U-Andrey Savchenko ovela egatsheni laseNizhny Novgorod leSikole Esiphakeme Sezomnotho ushicilele umphumela wocwaningo lwakhe emkhakheni wokufunda ngomshini ohlobene nokubona imizwa ebusweni babantu abakhona ezithombeni nakumavidiyo. Ikhodi ibhalwe nge-Python kusetshenziswa i-PyTorch futhi ilayisensi ngaphansi kwelayisensi ye-Apache 2.0. Amamodeli amaningana enziwe ngomumo ayatholakala, okuhlanganisa lawo alungele ukusetshenziswa kumadivayisi eselula.
Ngokusekelwe kumtapo wezincwadi, omunye umthuthukisi udale uhlelo lwe-sevimon, olukuvumela ukuthi ulandelele izinguquko emizweni usebenzisa ikhamera yevidiyo futhi usize ekulawuleni ukungezwani kwemisipha yobuso, isibonelo, ukuqeda ukucindezeleka ngokweqile, kuthinte ngokungaqondile isimo sengqondo futhi, ngokusetshenziswa isikhathi eside, vimbela ukubonakala kwemibimbi ebusweni. Ilabhulali ye-CenterFace isetshenziselwa ukunquma indawo yobuso kuvidiyo. Ikhodi ye-sevimon ibhalwe nge-Python futhi ilayisensi ngaphansi kwe-AGPLv3. Uma uyethula okokuqala, amamodeli ayalayishwa, emva kwalokho uhlelo aludingi uxhumano lwe-inthanethi futhi lusebenza ngokuzenzakalelayo. Imiyalo yokwethulwa ku-Linux/UNIX kanye ne-Windows isilungisiwe, kanye nesithombe sedokhu se-Linux.
I-Sevimon isebenza kanje: okokuqala, ubuso bubonakala esithombeni sekhamera, bese ubuso buqhathaniswa nemizwa eyisishiyagalombili (intukuthelo, ukwedelela, ukunengeka, ukwesaba, injabulo, ukuntula imizwelo, ukudabuka, ukumangala), ngemva kwalokho okunye amaphuzu afanayo anikezwa ngomzwelo ngamunye. Amanani atholiwe agcinwa kulogi ngefomethi yombhalo ukuze ahlaziywe ngokulandelayo uhlelo lwe-sevistat. Ngomuzwa ngamunye efayelini lezilungiselelo, ungasetha imingcele yamanani aphezulu naphansi, lapho weqa, isikhumbuzi sikhishwa ngokushesha.
Source: opennet.ru