Insufficient part segmentation accuracy when point cloud information is incomplete

[Jinbo-Zuo]

When I experiment with shapenet's full point cloud data, the segmentation accuracy is often as good as described in the paper. However, when I deployed the network to my project, I found that when the point cloud information is incomplete due to a single camera perspective, the segmentation effect is so poor that it cannot be used. Is there any way to improve the segmentation accuracy in the case of incomplete viewing angles? I haven't found any outstanding research in this area, maybe you have any engineering method to solve it?

Figure 1 is the part-segmentation effect when the full point cloud data of shapenet is used as input, and Figure 2 and Figure 3 are the part-segmentation results of images collected by a single camera.

[yan-margeo]

你好，我想请问一下您是怎么把部件分割可视化出来的