Summary of the 2022 Low-Power Deep Learning Semantic Segmentation Model Compression Competition for Traffic Scene In Asian Countries

The 2022 low-power deep learning semantic segmentation model compression competition for traffic scene in Asian countries held in IEEE ICME2022 Grand Challenges focuses on the semantic segmentation technologies in autonomous driving scenarios. The competition aims to semantically segment objects in traffic with low power and high mean intersection over union (mIOU) in the Asia countries (e.g., Taiwan), which contain several harsh driving environments. The target segmented objects include dashed white line, dashed yellow line, single white line, single yellow line, double dashed white line, double white line, double yellow line, main lane, and alter lane. There are 35,500 annotated images provided for model training revised from Berkeley Deep Drive 100K and 130 annotated images provided for example from Asian road conditions. Additional 2,012 testing images are used in the contest evaluation process, in which 1,200 of them are used in the qualification stage competition, and the rest are used in the final stage competition. There are in total 203 registered teams joining this competition, and the top 15 teams with the highest mIOU entered the final stage competition, from which 8 teams submitted the final results. The overall best model belongs to team “okt2077”, followed by team “asdggg” and team “AVCLab.” A special award for the best INT8 model development award is absent.