Skip to content

yanqiangmiffy/tibetan-mnist

Repository files navigation

tibetan

https://www.kesci.com/home/dataset/5bfe734a954d6e0010683839

https://www.kesci.com/home/project/5c063da7f55a5f002b73d159

模型测试

  • 3层全连接网络,两层Dropout

结果如下,在测试集准确率达到99.46%

loss: 0.0029 - acc: 0.9989 - val_loss: 0.0216 - val_acc: 0.9946

  • CNN

可以达到99.10%的准确率

loss: 0.0269 - acc: 0.9910 - val_loss: 0.0560 - val_acc: 0.9829

预测正确的情况:

预测错误的情况:

reference:https://www.one-tab.com/page/OeLJRrcFTW2DthBklKUMZw

keras webapp

在webapp目录下,运行train_lr.py和train_cnn.py训练模型,然后运行main.py启动web应用