It is reported that iDST, a core algorithm/data R&D team of Alibaba Group, ranked first in three road segmentation missions on KITTI platform, including two missions of particular scenario, UU_ROAD and UMM_ROAD with accuracy rate of 96.06% and 97.70% respectively and one evaluation mission of overall scenario, URBAN_ROAD with an accuracy rate of 96.76%. This is the first time for Alibaba to appear in road and scenario segmentation list of KITTI, the world's largest visual machine learning algorithm of autonomous driving.
KITTI dataset covers 5 major scenarios of urban areas, residential areas, roads, campuses and pedestrians. It adopts the assessment method of bird's eye view that is more fitting for autonomous vehicle.
Although most of autonomous driving research institutes are able to segment roads into several major pieces, there are still difficulties for them to recognize minor road boundaries. For example, when a vehicle pulls over, it is hard to identify the space between road edge and the vehicle according to the existing technologies. How to segment those road boundaries with more accuracy is the crucial problem researchers are trying to solve.
Alibaba announced that leading by Ren Xiaofeng, currently chief scientist and deputy dean at iDST, the R&D group came up with an Internet learning method based on online sample mining and added the modules of online data enhancement in network to improve the ability of network characterization and promotion. It may solve the problem caused by inaccurate segment of road verge and vehicle surroundings.
In addition, Alibaba’s iDST also appeared in other parts of this list, including vehicle detection, pedestrian detection and pedestrian re-identification. In accordance with relevant reports, these technologies are applied in the scenarios of Alibaba Cloud’s ET City Brain for the city transportation management.
In fact, Alibaba attaches great importance to its R&D abilities. In October, 2017, it announced that the company planned to invest more than RMB 100 billion over the next three years into a global research and development initiative called Alibaba DAMO Academy. Research will focus on a wide array of areas, including data intelligence, the Internet of Things, financial tech, quantum computing and human-machine interaction. Jack Ma, chairman of Alibaba Group, said that “DAMO Academy” is the best gift Alibaba presents to the world.