Chinese handwritten recognition

A demo of neural network based Chinese character recognition system with fixed point arithmetic.

Neural network architecture based on DenseNet
Three pretrained models(10MB each), each model train for different character sets, which are simplified Chinese and traditional Chinese
Trained on 1 billion+ characters with tons of fonts and data argumentation to increase accuracy of handwritten character recognition
Fixed point(Int32) computation
Heavy multithreading, no SIMD, should be as fast as NN library(GEMM) with SIMD support

GUI interface based on qt-simple-paint repo

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
chinese_ocr		chinese_ocr
screenshots		screenshots
.gitattributes		.gitattributes
LICENSE		LICENSE
README.md		README.md
canvaswidget.cpp		canvaswidget.cpp
canvaswidget.h		canvaswidget.h
chineseGeneralweights_quantized.c		chineseGeneralweights_quantized.c
main.cpp		main.cpp
mainwindow.cpp		mainwindow.cpp
mainwindow.h		mainwindow.h
mainwindow.ui		mainwindow.ui
simplePaint.pro		simplePaint.pro
simplifiedChineseBest_quantized.c		simplifiedChineseBest_quantized.c
traditionalChineseBest_quantized.c		traditionalChineseBest_quantized.c

Provide feedback