VENICE, Italy, October 24, 2017 /PRNewswire/ --
Today, Berkeley DeepDrive (BDD) and Nexar announced the release of 36,000 high frame-rate videos of driving, in addition to 5,000 pixel-level semantics-segmented labeled images, and invited public and private institution researchers to join the effort to develop accurate automotive perception and motion prediction models.
The BDD Industry Consortium, led by Professor Trevor Darrell of the Department of Electrical Engineering and Computer Sciences at UC Berkeley, investigates state-of-the-art technologies in computer vision and machine learning for automotive applications. Nexar, as a member of the consortium, contributes video and images captured by its road safety AI camera application deployed in over 100 countries worldwide.
Recently published BDD research using the Nexar driving data demonstrates autonomous driving algorithms that learn from large-scale video datasets (paper, video) and methods of cross-domain adaptation of semantic segmentation (FCNs in the Wild: Pixel-level Adversarial and Constraint-Based Adaptation, paper). More research in the area of autonomous driving, visual detection, and semantic segmentation using the driving datasets is expected to be published by the end of the year.
"We have consistently seen that quantity, quality, and a variety of data are key for producing research progress, as we've seen in areas of computer vision and machine learning," said Professor Alexei Efros of UC Berkeley, a leading computer vision researcher and BDD Principal Investigator who oversees the Berkeley-Nexar collaboration. "Training and validating models for real-world applications cannot achieve adequate results without a broad stream of annotated driving data--more pixels are the key."
In addition to the academic research, several groups have opened crowdsourced research challenges based on parts of the Nexar driving data. One example is the VisDA2017 challenge, which addresses the need for visual perception algorithms to perform well in domains not seen during training.
"Our challenge tests how well AI dashcam perception systems can perform on surprising visual domains, such as previously unseen geographic locations, weather conditions or types of terrain," said Professor Kate Saenko of Boston University, which was the VisDA2017 challenge organizer.
A second example is Nexar's Challenge-2, which focused on robust vehicle detection algorithms based on a diverse, global dataset. The winners of both challenges will be announced at ICCV 2017 on October 29th as part of the 4th TASK-CV workshop.
"We are pleased to be one of the BDD Industry Consortium's sponsors together with many other prominent members of the automotive industry," said Bruno Fernandez-Ruiz, Nexar Co-Founder and CTO. "The velocity of the development of end-to-end driving policies relies on an exponentially growing feed of diverse, real-world driving data to really cover the long tail of corner cases and rare scenarios involving multiple agents."
In addition to his role leading multiple research efforts at the UC Berkeley, Darrell also serves as Chief Scientist at Nexar. "I have been fortunate to work with leading researchers at Berkeley and Nexar, and see what great accomplishments arise when great data, great ideas, and great people are combined in both University and industrial research environments," he added.
Applying for access to the BDD data can be performed here, under applicable research and data privacy terms from Berkeley and subject to BDD's program discretion. All BDD sponsors have equal access to BDD data and code, which is also published openly for research use by academic and/or industrial researchers.
About Berkeley DeepDrive
The BDD Industry Consortium investigates state-of-the-art technologies in computer vision and machine learning for automotive applications. This multi-disciplinary center is housed at the University of California, Berkeley's Institute of Transportation Studies and is led by Professor Trevor Darrell, Faculty Director of Partners for Advanced Transportation Technology (PATH), with Professor Ken Goldberg and Thomas West serving as associate directors. The BDD consortium partners with private industry sponsors and brings faculty and researchers together from multiple departments and centers to develop new and emerging technologies with real-world applications in the automotive industry. Amongst its sponsors are leading firms such as Volkswagen, Ford, Toyota, Samsung, Bosch, Honda, Hyundai, Qualcomm and NVIDIA.
Nexar connects cars in the world's largest open vehicle-to-vehicle network. Leveraging connected smartphones and car cameras, Nexar provides real-time alerts to prevent vehicle, cyclist and pedestrian collisions.
Applying deep-learning and sensor fusion technologies over millions of car-sourced road miles, Nexar provides a series of data services for automakers, municipalities, and insurance carriers. For more information visit: http://www.getnexar.com.