> For the complete documentation index, see [llms.txt](https://nx.docs.scailable.net/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://nx.docs.scailable.net/ai-manager-v6.1.4/ai-manager-cloud-ui/upload-your-model/normalization.md).

# Normalization

Think of a digital image as a big grid filled with tiny colored dots, commonly called "pixels." Each pixel contains colors - usually a mix of red, green, and blue. These colors have values that range from 0 to 255. A value of 0 means there is none of that color in the pixel (it's totally off), and a value of 255 means that color is shining as brightly as possible.

To make it easier for a computer to analyze and compare different images, the color values can be normalized. One way to do this is by "normalizing" the color values in the image. Normalization is a term for adjusting these values so they fit within a new, consistent range, which helps in comparing images more fairly.

Here's how the normalization formula works:

* **normalized\_color\_value = (original\_color\_value - mean) / scale;**

In this formula:

* **original\_color\_value** is the initial value of the color (anywhere from 0 to 255).
* **mean** is the average of all the color values. Subtracting this mean helps center the color values around zero.
* **scale** is a number to divide by to keep the values within a new, smaller range. This could be something like the largest color difference or another predefined number.

For example:

* If the average (mean) color value is 100, and the scale is 50:
  * For a pixel with a red color value of 150:
    * Subtract the mean: `150 - 100 = 50`
    * Then divide by the scale: `50 / 50 = 1`
  * So, the normalized red value would be 1.

This process transforms the original color values to a new scale that's easier for the computer to work with, typically ranging between -1 and 1 or 0 and 1. This is comparable to converting measurements from different units into a single common unit for easier comparison.

In the AI Manager Cloud, set the normalization values as an integer array. For instance, for an RGB image, you might use \[123,234,242] for means and \[100,232,33] for scales. To leave the **original\_color\_value** unchanged, use 0 for mean and 1 for scale, that is, \[0,0,0] and \[1,1,1] for an RGB input.

<figure><img src="/files/LV7bjlYMsq4TZfjwoniK" alt="Add a model wizard showing the Normalization step with the Default preset applied,
       mean values [0,0,0] and standard deviation values [1,1,1]"><figcaption><p>The Normalization step in the upload wizard: the Default preset covers most ONNX models.</p></figcaption></figure>