Nexar issues automotive artificial intelligence challenge with global dataset
Nexar, a provider of a vehicle-to-vehicle (V2V) communication network for road safety, has challenged researchers to develop a geography-adaptive autonomous driving perception model. For the challenge, the company released its NEXET image dataset that reportedly includes more than 55,000 street-level images from more than 80 countries.
The goal of the challenge is to initiate a collaborative effort to address the problem of building a driving perception that performs consistently over different geographies. The key element for developing an all-weather, all-road, all-country driving perception is the ability to obtain data from a large and diverse training dataset. NEXET was curated to contain scenarios of varying lighting, weather, and topographical conditions as well as varying driving cultures in different countries to offer a comprehensive dataset.
"The robustness of learning driving policy models depends critically on having access to the largest possible training dataset exposing the true diversity of the 10 trillion miles that humans drive every year in the real world. Current approaches are trained using homogenous data from a small number of vehicles running in controlled environments, or in simulation, which fail to perform adequately in the true diversity of real-world dangerous corner cases," said Bruno Fernandez Ruiz, Cofounder and CTO of Nexar. "Safe driving requires continuously resolving a long tail of those corner cases. The only way to ensure safety in ADAS is to continuously capture as many of these cases as possible. By releasing this diverse dataset, we are opening our challenge to researchers to help us develop these algorithms and together create more robust ADAS models—essential to a safe autonomous future."
In this challenge, the task is to build a rear vehicle detector function that computes bounding boxes around each clearly visible vehicle in front. The detector should be looking for the vehicle(s) in front of the camera which are also driving in the same direction. The purpose of this task is to improve the forward vehicle collision warning feature, which requires an accurate bounding box around the rear side of the vehicle(s) ahead.
Researchers interested in joining the Nexar Challenge can find more details and sign up to receive the dataset here: https://www.getnexar.com/challenge-2/.