Unexpected Behavior for Segmentation Camera in Panoptic Mode With Nested Includes

[rsawtell]

Environment

• OS Version: Ubuntu 20.04
• Binary 6.9.0 (libignition-sensor6-segmentation-camera 6.3.0-2~focal)

The segmentation camera in panoptic mode returns strange results when the world includes a model that has further includes that are labeled. This includes the label image containing 0's instead of the expected label, and the color map showing the same color for copies of the same object.

Description

• Expected behavior: Objects have distinct colors and labels
• Actual behavior: Some objects have the same color, labels are missing or incorrect
• Behavior in semantic mode seems normal

Steps to reproduce

1. Copy the segmentation_camera.sdf to segmentation_test.sdf
2. Cut out all labeled objects except the first car
3. Paste the cut objects into a new file segmentation_includes.sdf wrapped with

<?xml version="1.0" ?>
<sdf version="1.6">
  <model name="segmentation_includes">
    <include>
      <name>Car2</name>
...
  </model>
</sdf>

4. put the segmentation_includes.sdf in a folder on the ignition path with a model.config to make it a valid model
5. include the model in the segmentation_test.sdf

    <include>
      <uri>model://segmentation_includes</uri>
    </include>

6. Run the modified demo: ign gazebo segmentation_test.sdf
7. press the play button to unpause the demo
8. edit the displayed images in the GUI to show the /panoptic/colored_map and /panoptic/labels_map
9. /panoptic/colored_map shows: the same color for the same objects except for the car that is still at the top level
10. /panoptic/labels_map: cones and cars have no labels

Output