Supplementary material : DeepCO 3 : Deep Instance Co-segmentation by Co-peak Search and Co-saliency Detection

We provide the class-wise statistics, including the number of images, the average number of instances per image, and the average area ratio of an instance to the whole image, of four collected datasets. Figure 1 gives the statistics for the COCO-VOC, VOC12, and SOC datasets, and Figure 2 reports for the COCO-NONVOC dataset. We randomly sample five images for each class from the four collected datasets and show them in Figure 3 (COCOVOC), Figure 4 (COCO-NONVOC) Figure 5 (VOC12), and Figure 6 (SOC).