MaSS13K: A Matting-level Semantic Segmentation Benchmark

Figure 1. Comparisons on the resolution and annotation quality between existing semantic segmentation datasets (a) COCO, (b) ADE20K, (c) EntitySeg and (d) our established dataset MaSS13K. Note that the 'others' category, which is highlighted in yellow, in our MaSS13K actually contains a group of well-segmented objects without specific class names. Please zoom-in for a better view.