Quality Evaluation Methods for Crowdsourced Image Segmentation

Abstract

Instance-level image segmentation provides rich information crucial for scene understanding in a variety of real-world applications. In this paper, we evaluate multiple crowdsourced algorithms for the image segmentation problem, including novel worker-aggregation-based methods and retrieval-based methods from prior work. We charac- terize the different types of worker errors observed in crowdsourced segmentation, and present a clustering algorithm as a preprocess- ing step that is able to capture and eliminate errors arising due to workers having different semantic perspectives. We demonstrate that aggregation-based algorithms attain higher accuracies than exist- ing retrieval-based approaches, while scaling better with increasing numbers of worker segmentations.