Mesh saliency via ranking unsalient patches in a descriptor space

Local-to-global mesh saliency

“…It also undesirably highlights the bottom edges of the girl bust and the neck of the Igea model. Similar appearance can be observed on the saliency maps produced by [32]. In general, it seems that their method prefers some local sharp edges such as the bottom edges of the girl bust and the neck of the Igea, the wings of the bird, the belly of the dolphin and the contour of the teddy.…”

“…Fig. 8 compares our results with those from [3,32,19,21,30]. The OutClass regression approach [3] fails to mark certain important features such as the eyes of the girl mesh.…”

“…In comparison, our method makes a better balance between local sharp features and globally salient features. For example, the head of the human model is not locally sharp and thus is not captured by [32]. But it is highlighted by our method since it is globally important.…”

“…In the local saliency map shown in (a), the head of the human model is not that salient compared with the feet and hands. This is a common problem for many other mesh saliency methods [19,30,32] where the tip of a long and thin protrusion has disproportionally high saliency. However, the research in [3] shows that the head or facial region of human should also be salient.…”

“…In order to examine the importance of the global saliency adjustment and demonstrate that it is the Fig. 8: Saliency detected by our method (first row) and the competing methods including (from the second row to the seventh row) [3], [32], [19], [21], [30] and [31]. Warm colors denote high saliency.…”