Tensorflow训练自己的手势图像集,从一开始损失值一直在降,但是训练精度一直保持不变。
我做了3类手势,每类手势有90张图,是28*28的灰度图,然后用tensorflow仿照MNIST的CNN网络做了一个架构,训练出来的效果就是:
Step=0, Train loss=1.1398,[Test accuracy=0.33]
Step=30, Train loss=0.7693,[Test accuracy=0.33]
Step=60, Train loss=0.4144,[Test accuracy=0.33]
Step=90, Train loss=0.1760,[Test accuracy=0.33]
Step=120, Train loss=0.0833,[Test accuracy=0.33]
Step=150, Train loss=0.0477,[Test accuracy=0.33]
Step=180, Train loss=0.0313,[Test accuracy=0.33]
Step=210, Train loss=0.0224,[Test accuracy=0.33]
Step=240, Train loss=0.0170,[Test accuracy=0.33]
Step=270, Train loss=0.0135,[Test accuracy=0.33]
Step=300, Train loss=0.0111,[Test accuracy=0.33]
Step=330, Train loss=0.0093,[Test accuracy=0.33]
Step=360, Train loss=0.0080,[Test accuracy=0.33]
Step=390, Train loss=0.0069,[Test accuracy=0.33]
………………
一直到Step=2000,Test accuracy都是0.33。。。
我才3类手势,随便猜一张图都是33%的概率,感觉这个训练完全没用啊……求解析,我不知道哪里出错了。但是同样的架构,用来识别它那个10个手写数字的精度就很好,我只是改了输入,输出端,还有一两个卷积核的大小。 蒙蔽ing