Sistimi e ncha ea ho lemoha mongolo oa EasyOCR

Morero Sesebelisoa sa EasyOCR Ho ntse ho ntlafatsoa mokhoa o mocha oa temoho ea mongolo o tšehetsang lipuo tse fetang 40, ho kenyelletsa Senyesemane, Sejeremane, Sefora, Sejapane, Sechaena, Sekorea, Uzbek, Azerbaijani le Lithuanian. Lipuo tse thehiloeng ho Cyrillic ha li so tšehetsoe, empa li ntse li eketsoa lethathamong la merero. Khoutu e ngotsoe ka Python ho sebelisoa moralo PyTorch и ajoa ke e nang le tumello tlas'a Apache 2.0. Bakeng sa ho kenya li fanoa Mefuta e lokiselitsoeng ea lipuo tse thehiloeng ho alfabeta ea Selatine le li-hieroglyphs.

Mekhoa ea ho ithuta ka mochini e sebelisoa ho khetholla le ho lemoha mongolo setšoantšong. Ho sebelisoa algorithm ea ho ithuta ka mochini ho khetholla mongolo Craft (Tlhokomeliso ea Libopeho tsa Sebaka Bakeng sa Mongolo) ho phethahatso bakeng sa PyTorch, e khonang ho hlakisa mongolo linthong tse ikhethileng, ho kenyeletsoa lileibole, matšoao a tlhahisoleseling le matšoao a tsela. Convolutional recurrent neural network e sebelisoa ho lemoha tatelano ea litlhaku CRNN (Convolutional Recurrent Neural Network, motsoako oa DCNN le RNN) le algorithm CTC BeamSearch CTC BeamSearch (Connectionist Temporal Classification) ho khetholla tlhahiso ea marang-rang ea neural ho hlahisa mongolo.

Source: opennet.ru

Eketsa ka tlhaloso