Baidu Apollo Releases the World's Largest Dataset for Self-Driving Cars

author: Eric Walz   

Baidu has announced the release of ‘Apollo Scape', billed as the world's largest open-source dataset for autonomous driving technology. The dataset is part of Baidu's open autonomous driving platform Apollo. The dataset eliminates the time it takes for manual data collection. Datasets are used in machine learning for regression and classification tasks.

Being open source, Baidu hopes its Apollo autonomous driving platform will become "the Android of the auto industry." Apollo gives developers access to a complete set of solutions and the source code, enabling a software engineer to convert a Lincoln MKZ or Ford Fusion into a self-driving vehicle in just 48 hours. ApolloScape's open sourced data now provides developers a comprehensive base for building self-driving vehicles.

According to Baidu, the data volume of ApolloScape is 10 times greater than any other open-source autonomous driving dataset, including Kitti and CityScapes. Kitti has been popular for many years. It is used by world-leading companies such as Baidu, Samsung and NVIDIA, as well as top universities, including Stanford and University of California.

This dataset can be utilized for perception, simulation scenes, road networks etc., as well as enabling autonomous vehicles to be trained in more complex urban driving environments, weather and traffic conditions.

Baidu said that ApolloScape will also conduct more research on the cutting-edge technology of simulation, the goal is to create the real world environment with the of the highest degree of detail.

Apollo Scape also defines 26 different semantic items including cars, bicycles, pedestrians, buildings, street lights, etc. — using a pixel-by-pixel semantic segmentation technique. The goal of image segmentation is to make them easier to analyze.

Apollo Scape can also simulate the complex scenario of dozens of vehicles driving on the same road. It is one of the most advanced intelligent driving simulation technologies available to help autonomous driving developers effectively examine and optimize forecasting, decision making and path planning.

The Apollo Scape dataset will save researchers and developers a huge amount of time on real-world sensor data collection. According to a Rand Corporation report, accumulating the same amount of real road test data from human drivers would require a fleet of 100 vehicles driving nonstop for 500 years.

Partnership with Berkeley's DeepDrive

In addition to the release of Apollo Scape, Baidu announced it has joined the Berkeley DeepDrive (BDD) Industry Consortium, a top-tier research alliance investigating state-of-the-art technologies in computer vision and machine learning for automotive applications.

Housed at the University of California, Berkeley and led by Professor Trevor Darrell, Faculty Director of the California Program for Advanced Transportation Technology (PATH), the BDD consortium has attracted big tech names as partners, including Ford, NVIDIA, Qualcomm, and General Motors.

BDD's main research focus is on the fields of artificial intelligence known as deep reinforcement learning, cross-modal transfer learning, and clockwork FCNs, used for camera images to reduce frame processing times, enabling faster video processing.

At the press conference, Baidu released a large number of automated driving open data sets for the Apollo platform. For autonomous driving developers, the high-quality real data is an indispensable "raw material." However, few companies have the ability to develop and maintain a suitable automated driving platform that regularly calibrates and collects vast amounts of new data.

Baidu vice president, chief of AI technology platform system (AIG), and Wang Haifeng, director of Baidu Research Institute, said: "Baidu and Berkeley's cooperation will rely on Apollo's open platform's industrial resources and Berkeley's top academic team to speed up automated driving. Technological innovation, theoretical innovation, and the process of landing applications."

"The partnership will incorporate Apollo's industrial resources and Berkeley's top academic team to ramp up the innovation of theoretical research, applied technology, and commercial applications."

Apollo Open Platform and BDD will jointly conduct a Workshop on Autonomous Driving at CVPR 2018 (IEEE International Conference on Computer Vision and Pattern Recognition) this June in Salt Lake City, where they will organize task competitions based on Apollo Scape.

Eric Walz

Originally hailing from New Jersey, Eric is a automotive & technology reporter covering the high-tech industry here in Silicon Valley. He has over 15 years of automotive experience and a bachelors degree in computer science. These skills, combined with technical writing and news reporting, allows him to fully understand and identify new and innovative technologies in the auto industry and beyond. He has worked at Uber on self-driving cars and as a technical writer, helping people to understand and work with technology.