Implementation:Kserve Kserve AnchorImages Explainer

Knowledge Sources	Kserve_Kserve
Domains	Explainability, Image Classification
Last Updated	2026-02-13 00:00 GMT

Overview

Concrete tool for generating anchor-based explanations for image classification models using the Alibi library provided by the KServe sample code.

Description

AnchorImages extends ExplainerWrapper and wraps the Alibi AnchorImage explainer. During construction, it receives a prediction function and a pre-built Alibi AnchorImage explainer instance. The explain() method converts input data to a NumPy array, checks whether the predictor returns class labels or probabilities (applying ArgmaxTransformer if needed to ensure class predictions), and then calls the Alibi anchor image explanation algorithm on the first image in the batch. The result identifies which image regions (superpixels) are sufficient for the classification decision.

Usage

Use this class as part of the Alibi explainer server pipeline when you need to generate visual explanations for image classification models, showing which regions of an image drive the model's prediction.

Code Reference

Source Location

• Repository: Kserve_Kserve
• File: docs/samples/explanation/alibi/alibiexplainer/alibiexplainer/anchor_images.py
• Lines: 1-49

Signature

class AnchorImages(ExplainerWrapper):
    def __init__(
        self,
        predict_fn: Callable,
        explainer: Optional[alibi.explainers.AnchorImage],
        **kwargs,
    ):
        ...

    def explain(self, inputs: List, headers: Dict[str, str] = None) -> Explanation:
        ...

Import

from alibiexplainer.anchor_images import AnchorImages

I/O Contract

Inputs

Constructor

Name	Type	Required	Description
predict_fn	Callable	Yes	Prediction function that accepts input arrays and returns predictions
explainer	Optional[alibi.explainers.AnchorImage]	Yes	A pre-built Alibi AnchorImage explainer instance (must not be None)
**kwargs	dict	No	Additional keyword arguments passed to the Alibi explain call

explain()

Name	Type	Required	Description
inputs	List	Yes	List of image arrays to explain (first image is used)
headers	Dict[str, str]	No	Optional HTTP headers

Outputs

explain()

Name	Type	Description
anchor_exp	Explanation	Alibi Explanation object containing anchor regions and metadata

Usage Examples

Basic Usage

from alibiexplainer.anchor_images import AnchorImages
import alibi
import numpy as np

# Assume a pre-built AnchorImage explainer and a predict function
def my_predict_fn(images):
    # Returns prediction probabilities
    return np.array([[0.1, 0.9]])

anchor_image_explainer = alibi.explainers.AnchorImage(predictor=my_predict_fn, image_shape=(28, 28, 3))
wrapper = AnchorImages(predict_fn=my_predict_fn, explainer=anchor_image_explainer)

# Generate an explanation for an image
sample_image = np.random.rand(1, 28, 28, 3).tolist()
explanation = wrapper.explain(sample_image)
print(explanation.anchor)

Related Pages

• Environment:Kserve_Kserve_Python_Runtime