A visual evaluation method based on the overlap ratio between superpixels and manually annotated regions

Superpixel segmentation, as a crucial step in image preprocessing, has been widely applied in visual tasks such as object segmentation and semantic segmentation. However, existing evaluation methods primarily rely on global quantitative metrics and lack intuitive visual analysis tools. To address this issue, this paper proposes a visual-assisted evaluation method based on the overlap ratio between superpixels and manually annotated regions. The proposed approach processes images and their corresponding annotations from a given dataset by assigning predefined labels and overlap thresholds. It automatically selects superpixels with high overlap ratios relative to target annotation regions and highlights these results on the original images using visualization tools. Experimental results demonstrate that the proposed method enables researchers to more intuitively observe the correspondence between superpixel segmentation and manual annotations, thereby providing effective support for the evaluation, adjustment, and improvement of superpixel algorithms.