AestheticScore<T>

Models & Types · AiDotNet.Metrics

Aesthetic Score metric using CLIP for evaluating image aesthetics.

How It Works

Aesthetic Score uses CLIP embeddings trained on aesthetic preference data to predict how visually appealing an image is. This is commonly used in image generation to filter or rank outputs by aesthetic quality.

Typical aesthetic scores (1-10 scale):

  • 7: High aesthetic quality

  • 5-7: Average aesthetic quality
  • <5: Low aesthetic quality

Constructors

Constructor Summary
AestheticScore(IMultimodalEmbedding<T> clipModel, Vector<T> aestheticWeights, string[] positivePrompts, string[] negativePrompts, T? aestheticBias) Initializes a new instance of AestheticScore calculator.

Methods

Method Summary
T Compute(double[] imageData) Computes aesthetic score for an image.
Vector<T> ComputeBatch(IList<double[]> images) Computes aesthetic scores for a batch of images.
int[] RankByAesthetic(IList<double[]> images) Ranks images by aesthetic score.

Fields

Field Summary
string[] DefaultNegativePrompts Default negative aesthetic prompts used for zero-shot aesthetic scoring.
string[] DefaultPositivePrompts Default positive aesthetic prompts used for zero-shot aesthetic scoring.