Using complex network communities to evaluate the correctness of object detection

Although today’s object detectors (ODs) are fairly powerful and advanced, they still suffer from an unneglectable level of error rate. In view of this issue, we develop our ConetCA framework to compute a relevance metric for an OD output, so that this metric serves as an indicator of how likely this OD output is faulty (i.e., containing misidentified objects). Our ConetCA framework is built upon the concept of complex networks and community clustering. We also performed an empirical study involving two open-source datasets and two deep-learning-based ODs. Our study results confirmed that the proposed relevance metric is highly effective.