TasksSotADatasetsPapersMethodsSubmitAbout
Papers With Code 2

A community resource for machine learning research: papers, code, benchmarks, and state-of-the-art results.

Explore

Notable BenchmarksAll SotADatasetsPapersMethods

Community

Submit ResultsAbout

Data sourced from the PWC Archive (CC-BY-SA 4.0). Built by the community, for the community.

Papers/Benchmarking Robustness in Object Detection: Autonomous Dr...

Benchmarking Robustness in Object Detection: Autonomous Driving when Winter is Coming

Claudio Michaelis, Benjamin Mitzkus, Robert Geirhos, Evgenia Rusak, Oliver Bringmann, Alexander S. Ecker, Matthias Bethge, Wieland Brendel

2019-07-17BenchmarkingData AugmentationAutonomous DrivingRobust Object DetectionInstance Segmentationobject-detectionObject Detection
PaperPDFCode(official)Code(official)Code(official)Code(official)

Abstract

The ability to detect objects regardless of image distortions or weather conditions is crucial for real-world applications of deep learning like autonomous driving. We here provide an easy-to-use benchmark to assess how object detection models perform when image quality degrades. The three resulting benchmark datasets, termed Pascal-C, Coco-C and Cityscapes-C, contain a large variety of image corruptions. We show that a range of standard object detection models suffer a severe performance loss on corrupted images (down to 30--60\% of the original performance). However, a simple data augmentation trick---stylizing the training images---leads to a substantial increase in robustness across corruption type, severity and dataset. We envision our comprehensive benchmark to track future progress towards building robust object detection models. Benchmark, code and data are publicly available.

Results

TaskDatasetMetricValueModel
Object DetectionPASCAL VOC 2007mPC [AP50]56.2Faster R-CNN with Stylized Training Data
Object DetectionPASCAL VOC 2007rPC [%]69.9Faster R-CNN with Stylized Training Data
Object DetectionPASCAL VOC 2007mPC [AP50]48.6Faster R-CNN
Object DetectionPASCAL VOC 2007rPC [%]60.4Faster R-CNN
Object DetectionCityscapesmPC [AP]17.2Stylized Training Data
Object DetectionCityscapes testmPC [AP]17.2Faster R-CNN with Stylized Training Data
Object DetectionCityscapes testrPC [%]47.4Faster R-CNN with Stylized Training Data
Object DetectionCityscapes testmPC [AP]12.2Faster R-CNN
Object DetectionCityscapes testrPC [%]33.4Faster R-CNN
Object DetectionCOCO (Common Objects in Context)mPC [AP]20.4Faster R-CNN with Stylized Training Data
Object DetectionCOCO (Common Objects in Context)rPC [%]58.9Faster R-CNN with Stylized Training Data
Object DetectionCOCO (Common Objects in Context)mPC [AP]18.2Faster R-CNN
Object DetectionCOCO (Common Objects in Context)rPC [%]50.2Faster R-CNN
3DPASCAL VOC 2007mPC [AP50]56.2Faster R-CNN with Stylized Training Data
3DPASCAL VOC 2007rPC [%]69.9Faster R-CNN with Stylized Training Data
3DPASCAL VOC 2007mPC [AP50]48.6Faster R-CNN
3DPASCAL VOC 2007rPC [%]60.4Faster R-CNN
3DCityscapesmPC [AP]17.2Stylized Training Data
3DCityscapes testmPC [AP]17.2Faster R-CNN with Stylized Training Data
3DCityscapes testrPC [%]47.4Faster R-CNN with Stylized Training Data
3DCityscapes testmPC [AP]12.2Faster R-CNN
3DCityscapes testrPC [%]33.4Faster R-CNN
3DCOCO (Common Objects in Context)mPC [AP]20.4Faster R-CNN with Stylized Training Data
3DCOCO (Common Objects in Context)rPC [%]58.9Faster R-CNN with Stylized Training Data
3DCOCO (Common Objects in Context)mPC [AP]18.2Faster R-CNN
3DCOCO (Common Objects in Context)rPC [%]50.2Faster R-CNN
2D ClassificationPASCAL VOC 2007mPC [AP50]56.2Faster R-CNN with Stylized Training Data
2D ClassificationPASCAL VOC 2007rPC [%]69.9Faster R-CNN with Stylized Training Data
2D ClassificationPASCAL VOC 2007mPC [AP50]48.6Faster R-CNN
2D ClassificationPASCAL VOC 2007rPC [%]60.4Faster R-CNN
2D ClassificationCityscapesmPC [AP]17.2Stylized Training Data
2D ClassificationCityscapes testmPC [AP]17.2Faster R-CNN with Stylized Training Data
2D ClassificationCityscapes testrPC [%]47.4Faster R-CNN with Stylized Training Data
2D ClassificationCityscapes testmPC [AP]12.2Faster R-CNN
2D ClassificationCityscapes testrPC [%]33.4Faster R-CNN
2D ClassificationCOCO (Common Objects in Context)mPC [AP]20.4Faster R-CNN with Stylized Training Data
2D ClassificationCOCO (Common Objects in Context)rPC [%]58.9Faster R-CNN with Stylized Training Data
2D ClassificationCOCO (Common Objects in Context)mPC [AP]18.2Faster R-CNN
2D ClassificationCOCO (Common Objects in Context)rPC [%]50.2Faster R-CNN
2D Object DetectionPASCAL VOC 2007mPC [AP50]56.2Faster R-CNN with Stylized Training Data
2D Object DetectionPASCAL VOC 2007rPC [%]69.9Faster R-CNN with Stylized Training Data
2D Object DetectionPASCAL VOC 2007mPC [AP50]48.6Faster R-CNN
2D Object DetectionPASCAL VOC 2007rPC [%]60.4Faster R-CNN
2D Object DetectionCityscapesmPC [AP]17.2Stylized Training Data
2D Object DetectionCityscapes testmPC [AP]17.2Faster R-CNN with Stylized Training Data
2D Object DetectionCityscapes testrPC [%]47.4Faster R-CNN with Stylized Training Data
2D Object DetectionCityscapes testmPC [AP]12.2Faster R-CNN
2D Object DetectionCityscapes testrPC [%]33.4Faster R-CNN
2D Object DetectionCOCO (Common Objects in Context)mPC [AP]20.4Faster R-CNN with Stylized Training Data
2D Object DetectionCOCO (Common Objects in Context)rPC [%]58.9Faster R-CNN with Stylized Training Data
2D Object DetectionCOCO (Common Objects in Context)mPC [AP]18.2Faster R-CNN
2D Object DetectionCOCO (Common Objects in Context)rPC [%]50.2Faster R-CNN
16kPASCAL VOC 2007mPC [AP50]56.2Faster R-CNN with Stylized Training Data
16kPASCAL VOC 2007rPC [%]69.9Faster R-CNN with Stylized Training Data
16kPASCAL VOC 2007mPC [AP50]48.6Faster R-CNN
16kPASCAL VOC 2007rPC [%]60.4Faster R-CNN
16kCityscapesmPC [AP]17.2Stylized Training Data
16kCityscapes testmPC [AP]17.2Faster R-CNN with Stylized Training Data
16kCityscapes testrPC [%]47.4Faster R-CNN with Stylized Training Data
16kCityscapes testmPC [AP]12.2Faster R-CNN
16kCityscapes testrPC [%]33.4Faster R-CNN
16kCOCO (Common Objects in Context)mPC [AP]20.4Faster R-CNN with Stylized Training Data
16kCOCO (Common Objects in Context)rPC [%]58.9Faster R-CNN with Stylized Training Data
16kCOCO (Common Objects in Context)mPC [AP]18.2Faster R-CNN
16kCOCO (Common Objects in Context)rPC [%]50.2Faster R-CNN

Related Papers

Visual Place Recognition for Large-Scale UAV Applications2025-07-20GEMINUS: Dual-aware Global and Scene-Adaptive Mixture-of-Experts for End-to-End Autonomous Driving2025-07-19AGENTS-LLM: Augmentative GENeration of Challenging Traffic Scenarios with an Agentic LLM Framework2025-07-18Training Transformers with Enforced Lipschitz Constants2025-07-17Disentangling coincident cell events using deep transfer learning and compressive sensing2025-07-17MUPAX: Multidimensional Problem Agnostic eXplainable AI2025-07-17Overview of the TalentCLEF 2025: Skill and Job Title Intelligence for Human Capital Management2025-07-17Pixel Perfect MegaMed: A Megapixel-Scale Vision-Language Foundation Model for Generating High Resolution Medical Images2025-07-17