The Baidu Apollo Autopilot Open Platform recently formally joined the DeepDrive Autodriving Industry Alliance at the University of California, Berkeley, and released the Apollo data and forward-looking technology brand Apollo Scape, officially opening ApolloScape’s large-scale automated driving data. Tencent According to science and technology reports, Apollo, Baidu's self-drive platform, officially joined UC Berkeley DeepDrive. Deep Learning Self Drive Industry Alliance UC Berkeley. UC Berkeley DeepDrive Deep Learning Autonomous Driving Industry Alliance research and application is mainly focused on computer vision and machine learning technology in the automotive field. Members include NVIDIA, Qualcomm, GM, Ford, and other 20 multinational corporations that develop self-driving businesses, research projects cover key automated technologies such as perception, planning and decision making, deep learning, etc. As Vice President of Baidu, AI Technology Platform Wang Haifeng, head of Baidu Research Institute, the head of the system (AIG), stated that the cooperation between Baidu and UC Berkeley will accelerate the technological theoretical innovation and landing process of self-drive through Apollo's open industry resources and UC Berkeley's top academic team. Baidu at the conference The announcement of ApolloScape's automatic driving open data information is the reason why the industry is rushing because the large amount of real data is an indispensable research raw material in autopilot development testing. However, few teams have the ability to develop and operate a The self-driving platform that regularly calibrates and collects new data. ApolloScape released by Apollo Open Platform not only has more than 10 times more data than Cityscapes' equivalent data sets, including perception, simulation scenarios, road network data, etc., hundreds of thousands of frames. Pixel semantically segmented high-resolution image data, further covering more complex environments, weather, traffic conditions, etc. In terms of data difficulty, ApolloScape data covers more complex road conditions, such as up to as much as a single image 162 vehicles or 80 pedestrians, while the open data set adopts the method of pixel-by-pixel semantic segmentation, is the most complex, most accurate, and most data-intensive autopilot dataset. The entire dataset published by ApolloScape contains Hundreds of thousands of frames of pixel-by-pixel semantic segmentation of high-resolution image data will facilitate research Personnel make full use of the data. Baidu defines 26 examples of different semantic items in the data set (including cars, bicycles, pedestrians, buildings, street lights, etc.). The future will further cover more complex environments, weather, and traffic conditions. ApolloScape will also conduct more forward-looking technology research on simulation, with the goal of creating the simulation platform with the highest degree of real-world reduction and the most abundant scene; At present, ApolloScape plans to invest dozens of autonomous vehicles simultaneously through the Apollo simulation platform. Running on a road network to simulate a real complex driving scenario is one of the most advanced smart driving simulation technologies available. It can help R&D personnel effectively check and optimize algorithms such as prediction, decision making and route planning, greatly improving the test diversity of automated driving. According to reports, the Apollo Open Platform will also co-host auto-driving seminars with UC Berkeley during the Computer Vision and Pattern Recognition Conference (CVPR) and define multiple task challenges with ApolloScape's large-scale data for global auto-driving developers and Researchers provide a common exploration of forward-looking areas of technological breakthroughs and application innovation New platform.
