ElasticTransform<T>

Models & Types · AiDotNet.Augmentation.Image

Applies elastic deformation to an image (Simard et al., 2003).

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

For Beginners

Imagine placing your image on a rubber sheet and randomly pushing and pulling different parts. The result looks naturally distorted, like how handwriting varies slightly each time you write the same letter. This is one of the most effective augmentations for digit/character recognition.

How It Works

Elastic deformation generates random displacement fields, smooths them with a Gaussian filter, and applies them to warp the image. This creates realistic local distortions similar to handwriting variations or biological tissue deformation.

When to use:

  • Handwritten digit/character recognition (MNIST, EMNIST)
  • Medical image segmentation (tissue deformation)
  • Any task where local shape variations are natural

Constructors

Constructor Summary
ElasticTransform(double alpha, double sigma, double fillValue, double probability) Creates a new elastic transform.

Properties

Property Summary
double Alpha { get; } Gets the alpha parameter controlling displacement magnitude.
double FillValue { get; } Gets the fill value for out-of-bounds pixels.
double Sigma { get; } Gets the sigma parameter controlling smoothness of the displacement field.

Methods

Method Summary
IDictionary<string, object> GetParameters()