Putting Image Manipulations in Context: Robustness Testing for Safe Perception

Pezzementi, Zachary; Tabor, Trenton; Yim, Samuel; Chang, Jonathan; Drozd, Bill; Guttendorf, David; Wagner, Michael; Koopman, Philip

doi:10.1109/ssrr.2018.8468619

Cited by 32 publications

(31 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…-Semantic analysis of triggering conditions: This property relates to the ability of the ML model to correctly process inputs from all relevant equivalence classes of the input domain. This may include ensuring a systematic coverage of an ontological definition of the operation domain as well as a focused search for corner cases, for example based on perturbations of inputs to adjust various properties that could trigger errors [23,25]. -Robustness: This property measures the resilience of an ML system to slight changes in the inputs.…”

Section: Category 2: Evaluating the Impact Of Insufficiencies In The ...mentioning

confidence: 99%

A causal model of safety assurance for machine learning

Burton¹

2022

Preprint

View full text Add to dashboard Cite

This paper proposes a framework based on a causal model of safety upon which effective safety assurance cases for ML-based applications can be built. In doing so, we build upon established principles of safety engineering as well as previous work on structuring assurance arguments for ML. The paper defines four categories of safety case evidence and a structured analysis approach within which these evidences can be effectively combined. Where appropriate, abstract formalisations of these contributions are used to illustrate the causalities they evaluate, their contributions to the safety argument and desirable properties of the evidences. Based on the proposed framework, progress in this area is re-evaluated and a set of future research directions proposed in order for tangible progress in this field to be made.

show abstract

Section: Category 2: Evaluating the Impact Of Insufficiencies In The ...mentioning

confidence: 99%

A causal model of safety assurance for machine learning

Burton¹

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…These solutions calculate a probability of a failure or a confidence score of the perception with a given frame of input data, without checking the perception results. The second direction is perception system validation and verification (Dreossi, Donzé, & Seshia, 2019;Pezzementi, Tabor, Yim, Chang, Drozd, Guttendorf, Wagner, & Koopman, 2018;. These solutions detect and identify failures during the validation and verification phase of developing the perception system.…”

Section: Introductionmentioning

confidence: 99%

Fault Diagnosis for the Perception Software in Highly Automated Vehicles and Reference Generation Based on World Model

2020

PHM_CONF

View full text Add to dashboard Cite

In a highly automated vehicle (HAV), the perception system performs critical sensing tasks such as object detection, scene understanding, etc. The perception software is based on complex intelligent algorithms and subjected to different failures such as missing object detection and false classification. It is a challenge to detect and identify these faults in the run time due to lack of ground truth and performance metrics. In this study, we introduce a method of perception software diagnostics, which is a generic method applicable to any state-of-the-art object detection models. In this method, perception results are compared with references generated with different sources, including pre-determined ground truth. The inconsistency between the perception results and the references, together with other performance metrics such as spatial variance, is used by a diagnostic logic to determine the fault type. Moreover, a method to generate references from a world model is presented. With validation, it is shown that the diagnostic algorithm provides a good performance in fault detection and isolation. This study enables fault mitigation in the run time and supports patch development in the development time.

show abstract

“…The sensitivity of computer vision algorithms to image quality KPIs was underlined recently “where performance drops catastrophically in response to barely perceptible changes” in automotive scenarios [ 30 ] as well as classification issues that can be induced by deliberately changing even single pixels [ 31 , 32 ].…”

Section: Introductionmentioning

confidence: 99%

Overview and Empirical Analysis of ISP Parameter Tuning for Visual Perception in Autonomous Driving

et al. 2019

View full text Add to dashboard Cite

Image quality is a well understood concept for human viewing applications, particularly in the multimedia space, but increasingly in an automotive context as well. The rise in prominence of autonomous driving and computer vision brings to the fore research in the area of the impact of image quality in camera perception for tasks such as recognition, localization and reconstruction. While the definition of “image quality” for computer vision may be ill-defined, what is clear is that the configuration of the image signal processing pipeline is the key factor in controlling the image quality for computer vision. This paper is partly review and partly positional with demonstration of several preliminary results promising for future research. As such, we give an overview of what is an

show abstract

Putting Image Manipulations in Context: Robustness Testing for Safe Perception

Cited by 32 publications

References 6 publications

A causal model of safety assurance for machine learning

A causal model of safety assurance for machine learning

Fault Diagnosis for the Perception Software in Highly Automated Vehicles and Reference Generation Based on World Model

Overview and Empirical Analysis of ISP Parameter Tuning for Visual Perception in Autonomous Driving

Contact Info

Product

Resources

About