Created by: yeyupiaoling

• 我使用LeNet-5或者使用VGG网络来做验证码识别时,通过训练后时预测的结果还是一样的,估计测试是随机的,训练根本不去作用
• 我是仿照book的第三章的例子,来识别验证码的灰度图,程序是没错的,不知道是哪里出了问题.
• 该项目的GitHub地址:https://github.com/yeyupiaoling/LearnPaddle/tree/master/note5 *以下是训练时输出的日志,可以看到Test的错误率一点都没有变,所以说训练的效果一点多没有用

Pass 0, Batch 0, Cost 61.418419, {'classification_error_evaluator': 0.96875}
...................................
Test with Pass 0, {'classification_error_evaluator': 0.959770143032074}

Pass 1, Batch 0, Cost 3.494206, {'classification_error_evaluator': 0.953125}
...................................
Test with Pass 1, {'classification_error_evaluator': 0.959770143032074}

Pass 2, Batch 0, Cost 3.490851, {'classification_error_evaluator': 0.9375}
...................................
Test with Pass 2, {'classification_error_evaluator': 0.959770143032074}

Pass 3, Batch 0, Cost 3.485399, {'classification_error_evaluator': 0.96875}
...................................
Test with Pass 3, {'classification_error_evaluator': 0.959770143032074}

Pass 4, Batch 0, Cost 3.489406, {'classification_error_evaluator': 0.9453125}
...................................
Test with Pass 4, {'classification_error_evaluator': 0.959770143032074}