How to choose scales for a new image at the input to the neural network?

Question

Neural networks

Let's say that there is a neural network that has already been "trained" and has created a file with class labels, which stores the most correct value given by the network and the class to which this value belongs.
Now we need to input (to get the prediction result) a new image, which was not in training. But what initial weights should be set for this image? What to be guided by?
PS as I understand it, by running a new image through the layers of the network, you can get the output value and compare it with the values ​​​​in the file with class labels. Does it actually work that way, or does it work differently?

Answer 1

"It" works differently.
After you have trained the network, no recalculations of the weights are carried out.
You feed a "new" image to the input (new - means one whose "correct label" you don't know!!!), you get the label assigned to it by the network as input. All. Problem solved.
And with what you are going to compare the received label - it is not clear from your text at all.