BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning

Yu, Fisher; Chen, Haofeng; Wang, Xin; Xian, Wenqi; Chen, Yingying; Liu, Fangchen; Madhavan, Vashisht; Darrell, Trevor

doi:10.1109/cvpr42600.2020.00271

Cited by 1,384 publications

(840 citation statements)

References 23 publications

Supporting

Mentioning

835

Contrasting

Unclassified

Order By: Relevance

“…In each image, at least 5 objects are detected. The ground truth correspondences are obtained through the Scalabel software [51]. 250 data instances are used for training and 250 instances are used for testing.…”

Section: Results In Real-world Furniture Assembly Scenariosmentioning

confidence: 99%

Bayesian Deep Graph Matching for Correspondence Identification in Collaborative Perception

Gao¹,

Zhang²

2021

Robotics: Science and Systems XVII

View full text Add to dashboard Cite

Correspondence identification is essential for multirobot collaborative perception, which aims to identify the same objects in order to ensure consistent references of the objects by a group of robots/agents in their own fields of view. Although recent deep learning methods have shown encouraging performance on correspondence identification, they suffer from two shortcomings, including the inability to address non-covisibility in collaborative perception that is caused by occlusion and limited fields of view of the agents, and the inability to quantify and reduce uncertainty to improve correspondence identification. To address both issues, we propose a novel uncertainty-aware deep graph matching method for correspondence identification in collaborative perception. Our new approach formulates correspondence identification as a deep graph matching problem, which identifies correspondences based upon graph representations that are constructed from the agents' observations. We introduce a novel deep graph matching network under the Bayesian framework to explicitly quantify uncertainty in the identified correspondences. In addition, we design a novel loss function that explicitly reduces correspondence uncertainty and perceptual non-covisibility during learning. We evaluate our approach in the robotics applications of collaborative assembly and multi-robot coordination using high-fidelity simulations and physical robots. Experiments have shown that, through addressing both uncertainty and non-covisibility, our approach achieves the state-of-the-art performance of correspondence identification.

show abstract

Section: Results In Real-world Furniture Assembly Scenariosmentioning

confidence: 99%

Bayesian Deep Graph Matching for Correspondence Identification in Collaborative Perception

Gao¹,

Zhang²

2021

Robotics: Science and Systems XVII

View full text Add to dashboard Cite

show abstract

“…StreetHazards contains artificially rendered driving scenes with inserted foreign objects. BDD-Anomaly also consists of driving scenes and was derived from the BDD100K dataset (Yu et al 2020) by selecting two classes as anomalous and removing images containing these classes from the training and validation sets.…”

Section: Segmentation Of Anomalous Regionsmentioning

confidence: 99%

The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection

et al. 2021

View full text Add to dashboard Cite

The detection of anomalous structures in natural image data is of utmost importance for numerous tasks in the field of computer vision. The development of methods for unsupervised anomaly detection requires data on which to train and evaluate new approaches and ideas. We introduce the MVTec anomaly detection dataset containing 5354 high-resolution color images of different object and texture categories. It contains normal, i.e., defect-free images intended for training and images with anomalies intended for testing. The anomalies manifest themselves in the form of over 70 different types of defects such as scratches, dents, contaminations, and various structural changes. In addition, we provide pixel-precise ground truth annotations for all anomalies. We conduct a thorough evaluation of current state-of-the-art unsupervised anomaly detection methods based on deep architectures such as convolutional autoencoders, generative adversarial networks, and feature descriptors using pretrained convolutional neural networks, as well as classical computer vision methods. We highlight the advantages and disadvantages of multiple performance metrics as well as threshold estimation techniques. This benchmark indicates that methods that leverage descriptors of pretrained networks outperform all other approaches and deep-learning-based generative models show considerable room for improvement.

show abstract

“…Although there are large-scale road datasets, such as KITTI [24], Cityscapes [25], Mapillary Vistas [26], ApolloScape [27] and BDD100k [28], the availability of useful image datasets for the foggy scenes evaluation is very low. Most of the existing datasets contain few or even no foggy scenes due to the difficulty of collecting and annotating them.…”

Section: Existing Fog Detection Datasetsmentioning

confidence: 99%

Sensors on the Move: Onboard Camera-Based Real-Time Traffic Alerts Paving the Way for Cooperative Roads

Iparraguirre

Amundarain

Brazalez

et al. 2021

Sensors

View full text Add to dashboard Cite

European road safety has improved greatly in recent decades. However, the current numbers are still far away to reach the European Commission’s road safety targets. In this context, Cooperative Intelligent Transport Systems (C-ITS) are expected to significantly improve road safety, traffic efficiency and comfort of driving, by helping the driver to make better decisions and adapt to the traffic situation. This paper puts forward two vision-based applications for traffic sign recognition (TSR) and real-time weather alerts, such as for fog-banks. These modules will support operators in road infrastructure maintenance tasks as well as drivers, giving them valuable information via C-ITS messages. Different state-of-the-art methods are analysed using both publicly available datasets (GTSB) as well as our own image databases (Ceit-TSR and Ceit-Foggy). The selected models for TSR implementation are based on Aggregated Chanel Features (ACF) and Convolutional Neural Networks (CNN) that reach more than 90% accuracy in real time. Regarding fog detection, an image feature extraction method on different colour spaces is proposed to differentiate sunny, cloudy and foggy scenes, as well as its visibility level. Both applications are already running in an onboard probe vehicle system.

show abstract

BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning

Cited by 1,384 publications

References 23 publications

Bayesian Deep Graph Matching for Correspondence Identification in Collaborative Perception

Bayesian Deep Graph Matching for Correspondence Identification in Collaborative Perception

The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection

Sensors on the Move: Onboard Camera-Based Real-Time Traffic Alerts Paving the Way for Cooperative Roads

Contact Info

Product

Resources

About