(Yicai Global) Nov. 2 -- China's software and information technology service company Iflytek Co. [SHE:002230] topped the list with an average accuracy of 81.4 percent, creating a new world record in Cityscapes Dataset for Semantic Urban Scene pixel-level image semantic segmentation evaluation in the field of international autonomous driving.
With the use of complex analysis technology, combined with the road scene analysis of special problems for algorithm migration, a multi-level semantic image segmentation scheme was built with a set of rich context information, hence, won the first place in the project evaluation, breaking a new world record, Iflytek said in a statement.
Pixel-level urban scene image semantic segmentation is known as the most accurate and most difficult algorithm for object recognition module in autonomous driving. The results of segmentation can provide more comprehensive road traffic reference information for vehicle computer, and provide guidance for path planning and related decision-making of high level autonomous driving in the future. Compared with the pedestrian detection and object recognition algorithms, the image semantic segmentation algorithm simplifies the overall decision process, and at the same time, provides more accurate decision-making reference for autonomous driving control, and reduces the accident rate.
Cityscapes Evaluation Dataset is up to now one of the most authoritative and professional image semantic segmentation evaluation datasets in the field of autonomous driving. It pays more attention to the urban road environment understanding in the real scene, and the task is more difficult and is closer to the hot demand of autonomous driving. Up to now, Cityscapes evaluation has attracted more than 40 teams for contest, including Alphabet Inc [NASDAQ:GOOG] (Google), Chinese University of Hong Kong and many other domestic and international outstanding innovative enterprises and top academic institutions.