Towards Robust CNN-based Object Detection through Augmentation with Synthetic Rain Variations

Volk, Georg; Bernuth, Alexander von; Hospach, Dennis; Bringmann, Oliver

doi:10.1109/itsc.2019.8917269

Cited by 57 publications

(23 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this work, we use most of the proposed image transformations and apply them to the Cityscapes dataset, PASCAL VOC 2012, and ADE20K (Cordts et al 2016;Everingham et al 2010;Zhou et al , 2016. Recent work deals further with model robustness against night images (Dai and Van Gool 2018), weather conditions (Sakaridis et al 2019(Sakaridis et al , 2018Volk et al 2019), and spatial transformations (Fawzi and Frossard 2015;Ruderman et al 2018). Zendel et al (2017) create a CV model for enabling to apply the hazard and operability analysis (HAZOP) to the computer vision domain and further provides an extensive checklist for image corruptions and visual hazards.…”

Section: Related Workmentioning

confidence: 99%

Benchmarking the Robustness of Semantic Segmentation Models with Respect to Common Corruptions

Kamann

Rother

2020

Int J Comput Vis

View full text Add to dashboard Cite

When designing a semantic segmentation model for a real-world application, such as autonomous driving, it is crucial to understand the robustness of the network with respect to a wide range of image corruptions. While there are recent robustness studies for full-image classification, we are the first to present an exhaustive study for semantic segmentation, based on many established neural network architectures. We utilize almost 400,000 images generated from the Cityscapes dataset, PASCAL VOC 2012, and ADE20K. Based on the benchmark study, we gain several new insights. Firstly, many networks perform well with respect to real-world image corruptions, such as a realistic PSF blur. Secondly, some architecture properties significantly affect robustness, such as a Dense Prediction Cell, designed to maximize performance on clean data only. Thirdly, the generalization capability of semantic segmentation models depends strongly on the type of image corruption. Models generalize well for image noise and image blur, however, not with respect to digitally corrupted data or weather corruptions.

show abstract

Section: Related Workmentioning

confidence: 99%

Benchmarking the Robustness of Semantic Segmentation Models with Respect to Common Corruptions

Kamann

Rother

2020

Int J Comput Vis

View full text Add to dashboard Cite

show abstract

“…They discover that adverse weather conditions such as heavy rain or fog degrade the detection performance of state-of-the-art object detectors. For example, a Faster RCNN reached a mAP approximately 3 % lower due to adverse weather [21]. Ren et al [22] in their study observe an even greater drop in performance of 27 %.…”

Section: Performance Influencing Meta-informationmentioning

confidence: 86%

“…The research of Michaelis et al [20] and Volk et al [21] focuses on the investigation of the influence of unfavorable weather conditions. They discover that adverse weather conditions such as heavy rain or fog degrade the detection performance of state-of-the-art object detectors.…”

Section: Performance Influencing Meta-informationmentioning

confidence: 99%

Identification and Explanation of Challenging Conditions for Camera-Based Object Detection of Automated Vehicles

Ponn

Kröger

Diermeyer

2020

Sensors

View full text Add to dashboard Cite

For a safe market launch of automated vehicles, the risks of the overall system as well as the sub-components must be efficiently identified and evaluated. This also includes camera-based object detection using artificial intelligence algorithms. It is trivial and explainable that due to the principle of the camera, performance depends highly on the environmental conditions and can be poor, for example in heavy fog. However, there are other factors influencing the performance of camera-based object detection, which will be comprehensively investigated for the first time in this paper. Furthermore, a precise modeling of the detection performance and the explanation of individual detection results is not possible due to the artificial intelligence based algorithms used. Therefore, a modeling approach based on the investigated influence factors is proposed and the newly developed SHapley Additive exPlanations (SHAP) approach is adopted to analyze and explain the detection performance of different object detection algorithms. The results show that many influence factors such as the relative rotation of an object towards the camera or the position of an object on the image have basically the same influence on the detection performance regardless of the detection algorithm used. In particular, the revealed weaknesses of the tested object detectors can be used to derive challenging and critical scenarios for the testing and type approval of automated vehicles.

show abstract

“…In the real world, such mismatches are commonly observed [35,25], and have led to significant performance drops in many deep learning algorithms [e.g., 9,37,46]. As a result, the reliability of current learning systems is substantially undermined in critical applications such as medical imaging [18,4], autonomous driving [20,70,6,63], and security systems [33].…”

Section: Introductionmentioning

confidence: 99%

OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalization

Ye¹,

Li²,

Bai³

et al. 2021

Preprint

View full text Add to dashboard Cite

Deep learning has achieved tremendous success with independent and identically distributed (i.i.d.) data. However, the performance of neural networks often degenerates drastically when encountering out-of-distribution (OoD) data, i.e., training and test data are sampled from different distributions. While a plethora of algorithms has been proposed to deal with OoD generalization, our understanding of the data used to train and evaluate these algorithms remains stagnant. In this work, we position existing datasets and algorithms from various research areas (e.g., domain generalization, stable learning, invariant risk minimization) seemingly unconnected into the same coherent picture. First, we identify and measure two distinct kinds of distribution shifts that are ubiquitous in various datasets. Next, we compare various OoD generalization algorithms with a new benchmark dominated by the two distribution shifts. Through extensive experiments, we show that existing OoD algorithms that outperform empirical risk minimization on one distribution shift usually have limitations on the other distribution shift. The new benchmark may serve as a strong foothold that can be resorted to by future OoD generalization research. * These authors contributed equally. 2 In the rest of this paper, we make no distinction between OoD generalization and DG (or between environments and domains), but the former is preferred for a greater audience.Preprint. Under review.

show abstract

Towards Robust CNN-based Object Detection through Augmentation with Synthetic Rain Variations

Cited by 57 publications

References 26 publications

Benchmarking the Robustness of Semantic Segmentation Models with Respect to Common Corruptions

Benchmarking the Robustness of Semantic Segmentation Models with Respect to Common Corruptions

Identification and Explanation of Challenging Conditions for Camera-Based Object Detection of Automated Vehicles

OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalization

Contact Info

Product

Resources

About