input data normalization

[Khalid Ashraf]

I am trying to port a trained network from an external DNN into Caffe. In order to do a one to one mapping at every layer output, I need to feed the same input
to both Caffe and external net. However, it looks like Caffe is normalizing the floating point values when the data layer is reading from leveldb. Is there a way to
turn off input data normalization ?

-khalid   

[Khalid Ashraf]

Got it, just put scale=1 in the prototxt file.

[dong zhihong]

stuck in the same problem for me, What the normalizing the floating point values mean? in prototxt file data layer have scale=1/256,  whats the meaning?

在 2014年10月14日星期二UTC+8上午11时26分56秒，Khalid Ashraf写道：

[Hong Cheng]

Hello，Do you know the reason now？Can you share it？

在 2014年10月25日星期六 UTC+9下午6:30:17，dong zhihong写道：