firenxygao / deblur

Dynamic Scene Deblurring with Parameter Selective Sharing and Nested Skip Connections
68 stars 11 forks source link

how to train the model? #5

Open c-dafan opened 4 years ago

c-dafan commented 4 years ago

Hello, I combine the code of this model with SRN-Deblur and train the model. But, the loss is very very large, and is easy to be Nan. So I want to know that how to train the model.

firenxygao commented 4 years ago

you can use truncated_normal_initializer with a small stddev value

c-dafan commented 4 years ago

you can use truncated_normal_initializer with a small stddev value

Thank you, it works.

ldlshizhu commented 3 years ago

you can use truncated_normal_initializer with a small stddev value

Thank you, it works.

你好~请问您能否分享一下结合的代码?十分感谢

zhaoguoqing12 commented 3 years ago

你可以使用 truncated_normal_initializer 和一个小的 stddev 值

但是看您并没有使用lstm类型训练吧 我有几个问题: 1.训练时你的loss也是巨大的吗? 2.只是更改了网络结构,其余训练参数没有改变是吗?