Hierarchical Intersection over Union Method: A Comprehensive Evaluation for Floor Plan Vectorization

Recent advancements in computer vision have significantly improved the performance of floor plan vectorization. However, achieving fair comparisons among different methods remains challenging because standard metrics often fail to capture the intrinsic characteristics of vector instances. Most studies rely on semantic segmentation metrics for quantitative assessment, but these metrics are insufficient for capturing detailed boundary aspects. This paper introduces the hierarchical intersection over union (HIoU) evaluation method to address this limitation. HIoU enhances compatibility across various vectorization approaches and strongly emphasizes boundary details, providing a more precise and well-directed standard for evaluation. Experimental results demonstrate the effectiveness of HIoU in delivering a comprehensive assessment, positioning it as a pivotal tool for advancing state-of-the-art vectorization tasks.