ImageTensor<T>

Models & Types · AiDotNet.Augmentation.Image

Represents an image as a tensor with image-specific metadata and operations.

For Beginners

An image on a computer is stored as numbers representing pixel colors. This class represents those numbers in a way that’s optimized for machine learning, while keeping track of important details like whether the image is in RGB or BGR format.

How It Works

ImageTensor wraps a Tensor to provide image-specific functionality:

  • Channel ordering (CHW vs HWC)
  • Color space awareness (RGB, BGR, HSV, etc.)
  • Normalization state tracking
  • Image-specific operations (crop, resize, color conversion)

Constructors

Constructor Summary
ImageTensor(Tensor<T> data, ChannelOrder channelOrder, ColorSpace colorSpace) Creates an ImageTensor from an existing tensor.
ImageTensor(int batchSize, int height, int width, int channels, ChannelOrder channelOrder, ColorSpace colorSpace) Creates a batched ImageTensor with specified dimensions.
ImageTensor(int height, int width, int channels, ChannelOrder channelOrder, ColorSpace colorSpace) Creates an ImageTensor with specified dimensions.

Properties

Property Summary
ChannelOrder ChannelOrder { get; set; } Gets or sets the channel ordering.
ColorSpace ColorSpace { get; set; } Gets or sets the color space.
IDictionary<string, object> Metadata { get; set; } Gets or sets additional metadata.
T[] NormalizationMean { get; set; } Gets or sets the normalization mean (per channel).
T[] NormalizationStd { get; set; } Gets or sets the normalization std (per channel).
Tensor<T> Data { get; } Gets the underlying tensor data.
ValueTuple<T, T>? OriginalRange { get; set; } Gets or sets the original value range before normalization.
bool IsNormalized { get; set; } Gets or sets whether the image is normalized to [0, 1].
int BatchSize { get; set; } Gets the batch size (1 for single images).
int Channels { get; set; } Gets the number of channels.
int Height { get; set; } Gets the image height in pixels.
int Width { get; set; } Gets the image width in pixels.

Methods

Method Summary
ImageTensor<T> Clone() Creates a deep copy of this image tensor.
ImageTensor<T> Crop(int x, int y, int width, int height) Extracts a rectangular region from the image.
ImageTensor<T> ToChannelOrder(ChannelOrder targetOrder) Converts this image to a different channel order.
T GetPixel(int y, int x, int channel) Gets a pixel value at the specified coordinates.
T[] GetPixelChannels(int y, int x) Gets all channel values at a pixel location.
int[] GetDimensions() Gets the tensor dimensions as an array.
void SetPixel(int y, int x, int channel, T value) Sets a pixel value at the specified coordinates.
void SetPixelChannels(int y, int x, T[] values) Sets all channel values at a pixel location.