A0711
Title: RankSEG: A consistent ranking-based framework for segmentation
Authors: Ben Dai - The Chinese University of Hong Kong (China) [presenting]
Chunlin Li - Iowa State University (United States)
Abstract: Segmentation has emerged as a fundamental field of computer vision and natural language processing, which assigns a label to every pixel/feature to extract regions of interest from an image/text. The Dice and IoU metrics are used to measure the degree of overlap between the ground truth and the predicted segmentation to evaluate the performance of segmentation. A theoretical foundation of segmentation is established with respect to the Dice/IoU metrics, including the Bayes rule and Dice-/IoU-calibration, analogous to classification-calibration or Fisher consistency in classification. The existing thresholding-based framework with most operating losses has been proven to be inconsistent with respect to the Dice/IoU metrics and thus may lead to a suboptimal solution. To address this pitfall, a novel consistent ranking-based framework, namely RankDice/RankIoU, is proposed, inspired by plug-in rules of the Bayes segmentation rule. Three numerical algorithms with GPU parallel execution are developed to implement the proposed framework in large-scale and high-dimensional segmentation. The numerical effectiveness of RankDice/mRankDice is demonstrated in various simulated examples and Fine-annotated CityScapes, Pascal VOC and Kvasir-SEG datasets with state-of-the-art deep learning architectures.
