Baidu has published a dataset called ApolloScape to foster advancements in autonomous vehicle research. The dataset includes 147,000 high resolution images with every pixel annotated to indicate whether it contains 26 different kinds of objects, such as buildings, vegetation, humans, and vehicles. This data includes scenes from widely varying driving conditions, which can allow researchers to simulate driving environments, improve object detection algorithms, and improve how autonomous vehicles navigate in poor weather. Baidu released ApolloScape as part of its forthcoming Workshop on Autonomous Driving Challenge, which aims to develop better computer vision algorithms for environmental perception.
Image: Kentaro IEMOTO.