Hi, i want to ask if this technique is normal,
Images for model normalized between -1,1 and labels for model normalized between 0,1.
I see this in some documentation and I am not sure if this is normal.
Hi,
The normalization for images is usually done by subtracting the mean and the dividing by the standard deviation. This helps the neurons convert faster.
So yeah, -1,1 is a typical behavior, as well as 0,1.
However, I’m not sure how or why the labels are normalized, could you provide more context?
This is for BatchNormalization, but it is the same principle
Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network
training by reducing internal covariate shift. 3 2015. doi: 1502.03167