Abstract
In this paper, we propose a novel evaluation metric for performance evaluation of semantic segmentation. In recent years, many studies have tried to train pixel-level classifiers on large-scale image datasets to perform accurate semantic segmentation. The goal of semantic segmentation is to assign a class label of each pixel in the scene. It has various potential applications in computer vision fields e.g., object detection, classification, scene understanding and Etc. To validate the proposed wIoU evaluation metric, we tested state-of-the art methods on public benchmark datasets (e.g., KITTI) based on the proposed wIoU metric and compared with other conventional evaluation metrics.
Abstract (translated)
URL
https://arxiv.org/abs/2107.09858
