PerturbedAttentionGuidance<T>

Models & Types · AiDotNet.Diffusion.Guidance

Perturbed Attention Guidance (PAG) for diffusion model inference.

Implements: IGuidanceMethod<T>

For Beginners

Instead of comparing “with prompt” vs “without prompt”, PAG compares “normal attention” vs “broken attention.” This gives better image quality, especially at high guidance scales where CFG can cause artifacts.

How It Works

PAG replaces the unconditional prediction with a “perturbed” prediction where self-attention maps are modified (e.g., replaced with identity). This produces better guidance than standard CFG without needing a separate unconditional pass.

Reference: Ahn et al., “Self-Rectifying Diffusion Sampling with Perturbed-Attention Guidance”, ECCV 2024

Constructors

Constructor Summary
PerturbedAttentionGuidance(double perturbationScale) Initializes a new Perturbed Attention Guidance instance.

Properties

Property Summary
GuidanceType GuidanceType { get; }

Methods

Method Summary
Tensor<T> Apply(Tensor<T> unconditional, Tensor<T> conditional, double scale, double timestep)