Baidu opens ApolloScape autonomous driving data set

Shanghai (Gasgoo)- On March 8 (the U.S. time), Baidu Apollo autonomous driving platform, officially joined Berkeley DeepDrive (BDD) Industry Consortium and opened its large-scale autonomous driving dataset, the ApolloScape. 

The BDD Industry Consortium, led by University of California, Berkeley, investigates cutting-edge technologies in computer vision and machine learning for automotive applications. Its participants include over 20 leading enterprises in autonomous driving realm worldwide, like NVIDIA, Qualcomm, General Motors and Ford Motor, etc. Besides, its R&D programs cover a number of key areas related to autonomous driving industry, such as perception, planning and decision making as well as deep learning.

Wang Haifeng, Baidu's vice president, head of AIG (Artificial Intelligence Group) and director of Baidu Research Institute, said that the cooperation between Baidu and BDD, relying on the industrialized resources from Apollo platform and leading academic team of BBD, will be devoted to accelerating the technological and theoretical innovation of autonomous driving industry and the realization of applications. 

At the conference, Baidu announced its autonomous driving dataset ApollpScape that includes a data amount 10 times larger than that of same-level datasets, like Cityscapes.

Huge amounts of data with high quality are indispensable to autonomous driving R&D. However, rare groups have such capability to develop and maintain an applicable autonomous driving platform that is able to regularly correct existing data and collect new data. 

The ApolloScape covers much more complicated road conditions. For example, 162 vehicles or 80 pedestrians simultaneously exist in a single image. Besides, the open dataset adopts the semantic segmentation to assign a label to each pixel in the image. Currently, ApolloScape is the dataset with the largest quantity, the most complicated scenarios and the most accurate labels worldwide.

Additionally, ApolloScape will carry out more cutting-edge technology researches about simulation with the objective of building a simulation platform with the highest reproduction degree and most abundant scenarios around the world. Based on Apollo's simulation platform, ApolloScape plans to let over 10 autonomous vehicles run in the same road network which can simulate the real-time complicated driving scenarios. This is one of the world's most advanced intelligent driving simulation technologies at present that can help developers examine and optimize the algorithms in respect of prediction, decision-making as well as road planning and enhance the diversity of autonomous driving tests.