Denormalize<T>

Models & Types · AiDotNet.Augmentation.Image

Reverses normalization of an image tensor, restoring original pixel value ranges.

Inheritance: AugmentationBase<T, ImageTensor<T>> → ImageAugmenterBase<T> → Denormalize<T>

For Beginners

After your model processes an image, the pixel values are in a normalized range (roughly -2 to +2). To display or save the image, you need to convert back to normal pixel values (0-255 or 0-1). This operation does that.

How It Works

Denormalization reverses the normalization operation: output = input * std + mean. This is the inverse of Normalize and restores pixel values to their original range for visualization or saving.

When to use:

  • Visualizing model inputs or intermediate activations
  • Saving processed images back to disk
  • Converting model output back to displayable format

Constructors

Constructor Summary
Denormalize(double[] mean, double[] std, bool clampOutput, double probability) Creates a new denormalization augmentation.

Properties

Property Summary
bool ClampOutput { get; } Gets whether to clamp output values to [0, 1].
double[] Mean { get; } Gets the per-channel mean values used in the original normalization.
double[] Std { get; } Gets the per-channel standard deviation values used in the original normalization.

Methods

Method Summary
Denormalize<T> Clip(bool clampOutput, double probability) Creates a denormalization for CLIP statistics.
Denormalize<T> ImageNet(bool clampOutput, double probability) Creates a denormalization for ImageNet statistics.
IDictionary<string, object> GetParameters()