Table of Contents

Class RealESRGAN<T>

Namespace
AiDotNet.Video
Assembly
AiDotNet.dll

Real-ESRGAN (Real Enhanced Super-Resolution GAN) for image and video super-resolution.

public class RealESRGAN<T> : NeuralNetworkBase<T>, INeuralNetworkModel<T>, INeuralNetwork<T>, IFullModel<T, Tensor<T>, Tensor<T>>, IModel<Tensor<T>, Tensor<T>, ModelMetadata<T>>, IModelSerializer, ICheckpointableModel, IParameterizable<T, Tensor<T>, Tensor<T>>, IFeatureAware, IFeatureImportance<T>, ICloneable<IFullModel<T, Tensor<T>, Tensor<T>>>, IGradientComputable<T, Tensor<T>, Tensor<T>>, IJitCompilable<T>, IInterpretableModel<T>, IInputGradientComputable<T>, IDisposable

Type Parameters

T

The numeric type used for calculations (typically float or double).

Inheritance
RealESRGAN<T>
Implements
IFullModel<T, Tensor<T>, Tensor<T>>
IModel<Tensor<T>, Tensor<T>, ModelMetadata<T>>
IParameterizable<T, Tensor<T>, Tensor<T>>
ICloneable<IFullModel<T, Tensor<T>, Tensor<T>>>
IGradientComputable<T, Tensor<T>, Tensor<T>>
Inherited Members
Extension Methods

Remarks

Real-ESRGAN is a practical super-resolution model that uses: - RRDB (Residual in Residual Dense Block) generator for deep feature extraction - U-Net discriminator for adversarial training - Combined loss: L1 (pixel) + Perceptual (VGG) + GAN (adversarial) - Second-order degradation model for realistic training data

For Beginners: Real-ESRGAN upscales images and video frames to higher resolution while adding realistic details. It's one of the most practical super-resolution models.

The network works by:

  1. Extracting deep features from low-resolution input using RRDB blocks
  2. Upsampling using pixel shuffle (efficient sub-pixel convolution)
  3. Training with adversarial loss to add realistic textures
  4. Using perceptual loss to ensure visual quality

Example usage:

// Create architectures
var generatorArch = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    inputHeight: 128, inputWidth: 128, inputDepth: 3);
var discriminatorArch = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    inputHeight: 512, inputWidth: 512, inputDepth: 3);

// Create model with 4x upscaling
var model = new RealESRGAN<double>(
    generatorArch, discriminatorArch,
    InputType.ThreeDimensional,
    scaleFactor: 4);

// Train
var (dLoss, gLoss) = model.TrainStep(lowResImages, highResTargets);

// Inference
var highRes = model.Upscale(lowResImage);

Reference: Wang et al., "Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data", ICCV 2021. https://arxiv.org/abs/2107.10833

Constructors

RealESRGAN(NeuralNetworkArchitecture<T>, NeuralNetworkArchitecture<T>, InputType, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>?, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>?, int, int, int, double, double, double, double)

Initializes a new instance of the Real-ESRGAN class.

public RealESRGAN(NeuralNetworkArchitecture<T> generatorArchitecture, NeuralNetworkArchitecture<T> discriminatorArchitecture, InputType inputType, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>? generatorOptimizer = null, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>? discriminatorOptimizer = null, int scaleFactor = 4, int numRRDBBlocks = 23, int numFeatures = 64, double residualScale = 0.2, double l1Lambda = 1, double perceptualLambda = 1, double ganLambda = 0.1)

Parameters

generatorArchitecture NeuralNetworkArchitecture<T>

Architecture for the RRDB-Net generator.

discriminatorArchitecture NeuralNetworkArchitecture<T>

Architecture for the U-Net discriminator.

inputType InputType

The type of input data (typically ThreeDimensional for images).

generatorOptimizer IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>

Optional optimizer for the generator. Default: Adam with lr=0.0001, beta2=0.99.

discriminatorOptimizer IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>>

Optional optimizer for the discriminator. Default: Adam with lr=0.0001, beta2=0.99.

scaleFactor int

Upscaling factor. Default: 4 (4x super-resolution).

numRRDBBlocks int

Number of RRDB blocks in generator. Default: 23 (Real-ESRGAN standard).

numFeatures int

Number of feature channels. Default: 64.

residualScale double

Residual scaling factor. Default: 0.2 (Real-ESRGAN standard).

l1Lambda double

L1 loss coefficient. Default: 1.0 (from Real-ESRGAN paper).

perceptualLambda double

Perceptual loss coefficient. Default: 1.0 (from Real-ESRGAN paper).

ganLambda double

GAN loss coefficient. Default: 0.1 (from Real-ESRGAN paper).

Remarks

For Beginners: Create a Real-ESRGAN model with sensible defaults: 

var model = new RealESRGAN<double>(generatorArch, discriminatorArch, InputType.ThreeDimensional);

Or customize for your needs:

// 2x upscaling with fewer blocks (faster but lower quality)
var model = new RealESRGAN<double>(
    generatorArch, discriminatorArch, InputType.ThreeDimensional,
    scaleFactor: 2, numRRDBBlocks: 16);

The default values are from the Real-ESRGAN paper and work well for most use cases.

RealESRGAN(NeuralNetworkArchitecture<T>, string, int)

Creates a Real-ESRGAN model using a pretrained ONNX model for inference.

public RealESRGAN(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, int scaleFactor = 4)

Parameters

architecture NeuralNetworkArchitecture<T>

The neural network architecture configuration.

onnxModelPath string

Path to the pretrained Real-ESRGAN ONNX model.

scaleFactor int

Upscaling factor of the pretrained model. Default: 4.

Remarks

For Beginners: Use this constructor when you have a pretrained Real-ESRGAN model in ONNX format. This is the fastest way to use Real-ESRGAN for inference without training.

Example:

var arch = new NeuralNetworkArchitecture<float>(
    inputType: InputType.ThreeDimensional,
    inputHeight: 128, inputWidth: 128, inputDepth: 3);
var model = new RealESRGAN<float>(arch, "realesrgan_x4.onnx");
var highRes = model.Upscale(lowResImage);

Note: Training is not supported in ONNX mode. Use the native constructor for training.

Pretrained ONNX models can be downloaded from: - https://github.com/xinntao/Real-ESRGAN (official) - Hugging Face model hub

Exceptions

FileNotFoundException

Thrown if the ONNX model file is not found.

Properties

Discriminator

Gets the U-Net discriminator network that judges image quality.

public UNetDiscriminator<T>? Discriminator { get; }

Property Value

UNetDiscriminator<T>

Remarks

Real-ESRGAN uses a U-Net discriminator that provides per-pixel feedback, which helps generate more detailed and realistic textures compared to standard discriminators that only provide a single real/fake score.

For Beginners: The discriminator learns to tell the difference between real high-resolution images and generated ones. This adversarial training helps the generator create more realistic outputs. Only available in native mode.

Generator

Gets the RRDB-Net generator network that produces super-resolved images.

public RRDBNetGenerator<T>? Generator { get; }

Property Value

RRDBNetGenerator<T>

Remarks

The generator uses Residual in Residual Dense Blocks (RRDB) for deep feature extraction, followed by pixel shuffle upsampling. This architecture is highly effective for super-resolution tasks.

For Beginners: The generator is the main network that takes your low-resolution image and outputs the high-resolution version. Only available in native mode.

LastDiscriminatorLoss

Gets the last discriminator loss value.

public T LastDiscriminatorLoss { get; }

Property Value

T

LastGeneratorLoss

Gets the last generator loss value.

public T LastGeneratorLoss { get; }

Property Value

T

ScaleFactor

Gets the upscaling factor for this model.

public int ScaleFactor { get; }

Property Value

int

SupportsTraining

Gets whether training is supported (only in native mode).

public override bool SupportsTraining { get; }

Property Value

bool

UseNativeMode

Gets whether this model uses native mode (true) or ONNX mode (false).

public bool UseNativeMode { get; }

Property Value

bool

Remarks

For Beginners: Native mode allows training and uses pure C# layers. ONNX mode loads a pre-trained model for inference only.

Methods

CreateNewInstance()

Creates a new instance of the same type as this neural network.

protected override IFullModel<T, Tensor<T>, Tensor<T>> CreateNewInstance()

Returns

IFullModel<T, Tensor<T>, Tensor<T>>

A new instance of the same neural network type.

Remarks

For Beginners: This creates a blank version of the same type of neural network.

It's used internally by methods like DeepCopy and Clone to create the right type of network before copying the data into it.

DeserializeNetworkSpecificData(BinaryReader)

Deserializes network-specific data that was not covered by the general deserialization process.

protected override void DeserializeNetworkSpecificData(BinaryReader reader)

Parameters

reader BinaryReader

The BinaryReader to read the data from.

Remarks

This method is called at the end of the general deserialization process to allow derived classes to read any additional data specific to their implementation.

For Beginners: Continuing the suitcase analogy, this is like unpacking that special compartment. After the main deserialization method has unpacked the common items (layers, parameters), this method allows each specific type of neural network to unpack its own unique items that were stored during serialization.

GetModelMetadata()

Gets the metadata for this neural network model.

public override ModelMetadata<T> GetModelMetadata()

Returns

ModelMetadata<T>

A ModelMetaData object containing information about the model.

InitializeLayers()

Initializes the layers of the neural network based on the architecture.

protected override void InitializeLayers()

Remarks

For Beginners: This method sets up all the layers in your neural network according to the architecture you've defined. It's like assembling the parts of your network before you can use it.

Predict(Tensor<T>)

Makes a prediction using the neural network.

public override Tensor<T> Predict(Tensor<T> input)

Parameters

input Tensor<T>

The input data to process.

Returns

Tensor<T>

The network's prediction.

Remarks

For Beginners: This is the main method you'll use to get results from your trained neural network. You provide some input data (like an image or text), and the network processes it through all its layers to produce an output (like a classification or prediction).

SerializeNetworkSpecificData(BinaryWriter)

Serializes network-specific data that is not covered by the general serialization process.

protected override void SerializeNetworkSpecificData(BinaryWriter writer)

Parameters

writer BinaryWriter

The BinaryWriter to write the data to.

Remarks

This method is called at the end of the general serialization process to allow derived classes to write any additional data specific to their implementation.

For Beginners: Think of this as packing a special compartment in your suitcase. While the main serialization method packs the common items (layers, parameters), this method allows each specific type of neural network to pack its own unique items that other networks might not have.

Train(Tensor<T>, Tensor<T>)

Trains the neural network on a single input-output pair.

public override void Train(Tensor<T> input, Tensor<T> expectedOutput)

Parameters

input Tensor<T>

The input data.

expectedOutput Tensor<T>

The expected output for the given input.

Remarks

This method performs one training step on the neural network using the provided input and expected output. It updates the network's parameters to reduce the error between the network's prediction and the expected output.

For Beginners: This is how your neural network learns. You provide: - An input (what the network should process) - The expected output (what the correct answer should be)

The network then:

  1. Makes a prediction based on the input
  2. Compares its prediction to the expected output
  3. Calculates how wrong it was (the loss)
  4. Adjusts its internal values to do better next time

After training, you can get the loss value using the GetLastLoss() method to see how well the network is learning.

TrainStep(Tensor<T>, Tensor<T>)

Performs one training step for Real-ESRGAN.

public (T discriminatorLoss, T generatorLoss) TrainStep(Tensor<T> lowResImages, Tensor<T> highResTargets)

Parameters

lowResImages Tensor<T>

Low-resolution input images.

highResTargets Tensor<T>

High-resolution target images.

Returns

(T Accuracy, T Loss)

Tuple of (discriminator loss, generator loss).

Remarks

This method performs one complete training iteration: 1. Generate super-resolved images from low-res input 2. Train discriminator to distinguish real from generated 3. Train generator to fool discriminator and minimize reconstruction loss

For Beginners: Call this method repeatedly during training: 

for (int epoch = 0; epoch < numEpochs; epoch++)
{
    foreach (var batch in dataLoader)
    {
        var (dLoss, gLoss) = model.TrainStep(batch.LowRes, batch.HighRes);
        Console.WriteLine($"D Loss: {dLoss:F4}, G Loss: {gLoss:F4}");
    }
}

UpdateParameters(Vector<T>)

Updates the network's parameters with new values.

public override void UpdateParameters(Vector<T> parameters)

Parameters

parameters Vector<T>

The new parameter values to set.

Remarks

For Beginners: During training, a neural network's internal values (parameters) get adjusted to improve its performance. This method allows you to update all those values at once by providing a complete set of new parameters.

This is typically used by optimization algorithms that calculate better parameter values based on training data.

Upscale(Tensor<T>)

Upscales a low-resolution image to high resolution.

public Tensor<T> Upscale(Tensor<T> lowResImage)

Parameters

lowResImage Tensor<T>

The low-resolution input image tensor.

Returns

Tensor<T>

The super-resolved high-resolution image tensor.

Remarks

For Beginners: Use this method after training to upscale images: 

var highResImage = model.Upscale(lowResImage);
Edit this page