PASDModel<T>

Models & Types · AiDotNet.Diffusion.SuperResolution

PASD: Pixel-Aware Stable Diffusion for real-world image super-resolution.

Inheritance: DiffusionModelBase<T> → LatentDiffusionModelBase<T> → PASDModel<T>

For Beginners

When upscaling images, it’s crucial to keep the original structure intact while adding detail. PASD weaves the low-resolution image information directly into the diffusion model’s attention mechanism, ensuring the output matches the input’s structure perfectly while adding sharp, realistic details.

How It Works

PASD uses pixel-aware cross-attention to inject low-resolution structural information directly into the diffusion U-Net’s attention layers. This ensures the super-resolved image preserves the original structure while generating realistic high-frequency details.

Technical specifications:

  • Architecture: SD2.1 U-Net with pixel-aware cross-attention
  • Text encoder: OpenCLIP ViT-H/14 (1024-dim) — uses image captions for guidance
  • Pixel-aware attention: LR features injected into cross-attention at every layer
  • Scale factor: 4x upscaling
  • Personalized stylization also supported

Reference: Yang et al., “Pixel-Aware Stable Diffusion for Realistic Image Super-resolution and Personalized Stylization”, 2024

Example

var options = new DiffusionModelOptions<float> { LatentChannels = 4, DefaultInferenceSteps = 20 };
var lowRes = Tensor<float>.CreateRandom(new[] { 1, 4, 32, 32 });
var trainX = Tensor<float>.CreateRandom(4, 8);
var trainY = Tensor<float>.CreateRandom(4, 2);
var result = new AiModelBuilder<float, Tensor<float>, Tensor<float>>()
    .ConfigureModel(new PASDModel<float>(options: options))
    .Build(trainX, trainY);
var highRes = result.Predict(lowRes);

Constructors

Constructor Summary
PASDModel(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, UNetNoisePredictor<T> predictor, StandardVAE<T> vae, IConditioningModule<T> conditioner, int? seed)

Properties

Property Summary
IConditioningModule<T> Conditioner { get; }
INoisePredictor<T> NoisePredictor { get; }
IVAEModel<T> VAE { get; }
int LatentChannels { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()