project A new optical text recognition system iri kugadzirwa inotsigira mitauro inodarika makumi mana, inosanganisira Chirungu, ChiGerman, French, Japanese, Chinese, Korean, Uzbek, Azerbaijani neLithuanian. Mitauro yeCyrillic-yakavakirwa haisati yatsigirwa, asi iri kuwedzerwa kune runyorwa rwezvirongwa. Iyo kodhi yakanyorwa muPython uchishandisa chimiro ΠΈ ine rezinesi pasi peApache 2.0. Zvekurodha akagadzirira-akagadzirwa modhi yemitauro yakavakirwa pachiLatin alphabet uye hieroglyphs.
Nzira dzekudzidza dzemuchina dzinoshandiswa kuona uye kuziva zvinyorwa mumufananidzo. Muchina wekudzidza algorithm unoshandiswa kuona zvinyorwa (Kuzivisa Hunhu-Nzvimbo Yezvinyorwa) mu yePyTorch, inokwanisa kuratidza zvinyorwa pazvinhu zvisina tsarukano, zvinosanganisira mavara, zviratidzo zveruzivo uye zviratidzo zvemumugwagwa. A convolutional recurrent neural network inoshandiswa kuona maitiro akatevedzana (Convolutional Recurrent Neural Network, musanganiswa weDCNN neRNN) uye algorithm CTC BeamSearch (Connectionist Temporal Classification) yekusarudza neural network inobuda mukumiririra mavara.
Source: opennet.ru
