Sabon tsarin gane rubutu na gani EasyOCR

Aikin EasyOCR Ana samar da sabon tsarin gane rubutu na gani wanda ke tallafawa fiye da harsuna 40, ciki har da Ingilishi, Jamusanci, Faransanci, Jafananci, Sinanci, Koriya, Uzbek, Azerbaijan da Lithuanian. Har yanzu ba a tallafawa harsunan tushen Cyrillic ba, amma ana ƙara su cikin jerin tsare-tsare. An rubuta lambar a cikin Python ta amfani da tsarin PyTorch и rarraba ta lasisi a ƙarƙashin Apache 2.0. Don lodawa ana bayar da su shirye-shiryen da aka yi don harsuna dangane da haruffan Latin da hieroglyphs.

Ana amfani da hanyoyin koyan inji don ganowa da gane rubutu a hoto. Ana amfani da algorithm koyan inji don gano rubutu SANA'A (Fadar Halaye-Yanki Don Rubutu) in aiwatarwa don PyTorch, mai ikon nuna rubutu akan abubuwa na sabani, gami da alamomi, alamun bayanai da alamun hanya. Ana amfani da hanyar sadarwa mai maimaita juzu'i don gane jerin halaye CRNN (Convolutional Recurrent Neural Network, hade da DCNN da RNN) da kuma algorithm CTC BeamSearch CTC BeamSearch (Rarrabi na ɗan lokaci mai haɗin gwiwa) don ƙaddamar da fitarwar cibiyar sadarwar jijiyoyi zuwa wakilcin rubutu.

source: budenet.ru

Add a comment