Skip to content

Image classification explainer

Available

ImageClassificationExplainer is implemented and has a complete usage guide.

The image-classification explainer perturbs image regions, observes changes in model predictions, and estimates which regions have the strongest local influence.

Current documented capabilities include:

  • built-in PyTorch classification models;
  • custom PyTorch models and preprocessing;
  • DINOv2 image embeddings;
  • supported segmentation models or a supplied mask;
  • multiple statistical distance measures;
  • image and image-heatmap visualisations.

Read the complete image-classification tutorial

Open the image explainer API reference

Scope

The currently implemented workflow is image classification. Broader image tasks should be described as planned until a corresponding implementation and test suite are added.