High Efficiency Dataset Generation for Semantic Video Segmentation on Road Intersection
Wataru Nagai, Takafumi Katayama, Tian Song, Takashi Shimamoto · 2022
This work proposes a highly efficient dataset generation method for semantic video segmentation at road intersections. Semantic segmentation is an excellent method to determine the class for each pixel in an image. However, semantic segmentation requires the human cost of generating a dataset with annotations for each pixel. For this reason, the dataset must be generated in an ingenious way. The dataset is dedicated to road intersections that have complex situations and high accident rates. The simulation results show that the proposed method can achieve the fast generation of the dataset. Additionally, the proposed segmentation performance can be achieved with the easy generation method of the dataset.