Metrics for Saliency Map Evaluation of Deep Learning Explanation Methods

Gomez, Tristan; Fréour, Thomas; Mouchère, Harold

doi:10.1007/978-3-031-09037-0_8

Cited by 21 publications

(22 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this case, to evaluate an explainer, human judgment or GFDMs, that is "a reference" is not needed: we can call the metrics proposed in this case as no-reference. One group of these methods constitute the fidelity metrics [17]. These evaluation methods are based on the principle that if the perturbations are induced in the parts of input data highlighted as important by explainers, then the classification score will change in which case the explainer can be considered as good.…”

Section: State-of-the-art In Evaluation Of Explainers For Cnn Classif...mentioning

confidence: 99%

“…All of them are based on the changes of the class score for a given image after it has been modified accordingly to the importance of pixels in explanation maps. In [17], the authors criticize DAUC and IAUC for the fact that they use only rank of the score and not its value, and propose Deletion Correlation (DC) and Insertion Correlation (IC) metrics. They are computed as correlation coefficient between the difference of scores due to masking or adding details and the score of saliency of pixels masked/added progressively.…”

Section: State-of-the-art In Evaluation Of Explainers For Cnn Classif...mentioning

confidence: 99%

“…The metric Sparsity [17], qualifies the distribution of importance scores in the explanation maps per se. Its authors claim that higher sparsity makes the map more interpretable by humans, as it is concentrated on a small amount of elements.…”

Section: State-of-the-art In Evaluation Of Explainers For Cnn Classif...mentioning

confidence: 99%

See 2 more Smart Citations

Evaluation of Explanation Methods of AI - CNNs in Image Classification Tasks with Reference-based and No-reference Metrics

Zhukov¹,

Benois‐Pineau²,

Giot³

2023

AAIML

View full text Add to dashboard Cite

The most popular methods in AI-machine learning paradigm are mainly black boxes. This is why explanation of AI decisions is of emergency. Although dedicated explanation tools have been massively developed, the evaluation of their quality remains an open research question. In this paper, we generalize the methodologies of evaluation of post-hoc explainers of CNNs’ decisions in visual classification tasks with reference and no-reference based metrics. We apply them on our previously developed explainers (FEM1 , MLFEM), and popular Grad-CAM. The reference-based metrics are Pearson correlation coefficient and Similarity computed between the explanation map and its ground truth represented by a Gaze Fixation Density Map obtained with a psycho-visual experiment. As a no-reference metric, we use stability metric, proposed by Alvarez-Melis and Jaakkola. We study its behaviour, consensus with reference-based metrics and show that in case of several kinds of degradation on input images, this metric is in agreement with reference-based ones. Therefore, it can be used for evaluation of the quality of explainers when the ground truth is not available.

show abstract

Section: State-of-the-art In Evaluation Of Explainers For Cnn Classif...mentioning

confidence: 99%

Section: State-of-the-art In Evaluation Of Explainers For Cnn Classif...mentioning

confidence: 99%

See 1 more Smart Citation

Evaluation of Explanation Methods of AI - CNNs in Image Classification Tasks with Reference-based and No-reference Metrics

Zhukov¹,

Benois‐Pineau²,

Giot³

2023

AAIML

View full text Add to dashboard Cite

show abstract

“…Maximizing this metric corresponds to an improvement. Like DAUC, the deletion correlation (DC) 64 metric also consists in gradually masking the input image by following the order suggested by the saliency map, but instead of the AUC of the score curve, it is defined as the linear correlation of the class score variations and the saliency scores. As it measures the correlation between the saliency of a pixel and its impact on the class score, maximizing this metric is an improvement.…”

Section: Multi-step Metricsmentioning

confidence: 99%

Computing and evaluating saliency maps for image classification: a tutorial

Gomez

Mouchère

2023

J. Electron. Imag.

Self Cite

View full text Add to dashboard Cite

Facing the black-box nature of deep learning models for image classification, a popular trend in the literature proposes methods to generate explanations in the form of heat maps indicating the areas that played an important role in the models' decisions. Such explanations are called saliency maps and constitute an active field of research, given that many fundamental questions are yet to be answered: how to compute them efficiently? How to evaluate them? What exactly can they be used for? Given the increasing rate at which papers are produced and the vast amount of literature that is already existing, we propose our study to help newcomers become part of this community and to contribute to the research field. First, the two existing approaches to generate saliency maps are discussed, namely post-hoc methods and attention models. Post-hoc methods are generic algorithms that can be applied to any model from a given class without requiring fine-tuning. On the contrary, attention models are ad-hoc architectures that generate a saliency map during the inference phase to guide the decision. We show that both approaches can be divided into several subcategories and illustrate each of them with one important model or method. Second, we present the current methodologies used to evaluate saliency maps, including objective and subjective protocols, depending on whether or not they involve users. Among objective methods, we notably detail faithfulness metrics and propose an implementation featuring the faithfulness metrics discussed in this paper (https://github. com/TristanGomez44/metrics-saliency-maps).

show abstract

“…Gomez et. al created a deep network set with the purpose of constructing a compact and uniform saliency map [102]. This map differentiates pixels from the object boundary.…”

Section: Cnn Applications For Specialized Object Detectionmentioning

confidence: 99%

An Evaluation of Deep Learning-Based Object Identification

Kolluri

Das

2022

IJRITCC

View full text Add to dashboard Cite

Identification of instances of semantic objects of a particular class, which has been heavily incorporated in people's lives through applications like autonomous driving and security monitoring, is one of the most crucial and challenging areas of computer vision. Recent developments in deep learning networks for detection have improved object detector accuracy. To provide a detailed review of the current state of object detection pipelines, we begin by analyzing the methodologies employed by classical detection models and providing the benchmark datasets used in this study. After that, we'll have a look at the one- and two-stage detectors in detail, before concluding with a summary of several object detection approaches. In addition, we provide a list of both old and new apps. It's not just a single branch of object detection that is examined. Finally, we look at how to utilize various object detection algorithms to create a system that is both efficient and effective. and identify a number of emerging patterns in order to better understand the using the most recent algorithms and doing more study.

show abstract

Metrics for Saliency Map Evaluation of Deep Learning Explanation Methods

Cited by 21 publications

References 15 publications

Evaluation of Explanation Methods of AI - CNNs in Image Classification Tasks with Reference-based and No-reference Metrics

Evaluation of Explanation Methods of AI - CNNs in Image Classification Tasks with Reference-based and No-reference Metrics

Computing and evaluating saliency maps for image classification: a tutorial

An Evaluation of Deep Learning-Based Object Identification

Contact Info

Product

Resources

About