OpenVINO™ Explainable AI Toolkit (2/3): Deep Dive

This is the second notebook in series of exploring OpenVINO™ Explainable AI (XAI):

1. OpenVINO™ Explainable AI Toolkit (1/3): Basic
2. OpenVINO™ Explainable AI Toolkit (2/3): Deep Dive
3. OpenVINO™ Explainable AI Toolkit (3/3): Saliency map interpretation

OpenVINO™ Explainable AI (XAI) provides a suite of XAI algorithms for visual explanation of OpenVINO™ Intermediate Representation (IR) models.

Using OpenVINO XAI, you can generate saliency maps that highlight regions of interest in input images from the model's perspective. This helps users understand why complex AI models produce specific responses.

This notebook shows an example how to use OpenVINO XAI.

It depicts a heatmap with areas of interest where neural network (classification or detection) focuses before making a decision.

Example: Saliency map for flat-coated retriever class for MobileNetV3 classification model:

Notebook Contents

The tutorial consists of the following steps:

• Run Explainer in AUTO mode
• Specify preprocess and postprocess functions
• Run Explainer in WHITEBOX mode
  • Insert XAI branch to IR or PyTorch model to use updated model in own pipelines
• Run Explainer in BLACKBOX mode
• Advanced: add label names and use them to save saliency maps instead of label indexes

These are explainable AI algorithms supported by OpenVINO XAI:

Domain	Task	Type	Algorithm	Links
Computer Vision	Image Classification	White-Box	ReciproCAM	arxiv / src
			VITReciproCAM	arxiv / src
			ActivationMap	experimental / src
		Black-Box	AISEClassification	src
			RISE	arxiv / src
	Object Detection	White-Box	ClassProbabilityMap	experimental / src
		Black-Box	AISEDetection	src

Installation Instructions

This is a self-contained example that relies solely on its own code.

We recommend running the notebook in a virtual environment. You only need a Jupyter server to start. For details, please refer to Installation Guide.