查看文章

Answers. The ground-truth point maps of the IOC examples shown in the introduction of the main paper are presented in Fig. 1. For a normal visual system, it is difficult to find all the object (fish) instances in the given examples because some objects are highly blended in their environment. Surprisingly, there are a total of 20 and 61 objects in the two examples, respectively. More examples. We show more example images of IOC-fish5K in Fig. 2. Our dataset contains plenty of high-quality images with various indiscernible levels and object densities. It would contribute to research areas such as indiscernible scene understanding and dense object counting. Annotation details. As mentioned in our main paper, the annotation process is split into three steps. For the first step, we studied books and watched Youtube videos to learn the background knowledge about sea animals before instructing annotators. The used books include Wonders of the Red