Revision history [back]

We're working on a system that does something like this (for large environments), but it's not released yet.

Until then, you could use the point clouds extracted by your segmenter as a source of data for labeling. For example, you could compute the mean color of the cloud, and then bucket that into "red", "blue", "orange", or whatever. Computing a "can" label could be done using cylinder fitting; computing "metallic" would be quite a bit trickier.