Major issues with Unity samples — ZED SDK 5.0: need a clear status update

Hello Stereolabs team,

I’m opening this topic to highlight several blocking issues I’ve encountered while testing the latest ZED SDK 5.0 and ZED Unity Plugin 5.0.1, whether using the official GitHub repo or by importing the package into a fresh Unity project.

Test Environment

  • Unity: 2022.3.29 (LTS) and 6000.0.33
  • ZED SDK: 5.0
  • Unity ZED SDK: 5.0.1 — GitHub link
  • Tested scenes: Assembly Definition, BodyTracking, ObjectDetection, ObjectPlacement, PointCloud (official samples)

Issues Encountered

BodyTracking

  • Scene: BodyTrackingMulti/BodyTrackingSingle → Depth occlusion problem (already reported in another topic).
  • Advanced: Mask Bodytracking
    • ZEDCommon uses System.IntPtr mask
    • DetectBody.CS: ZEDMatToTexture_CPU() + Marshal.Copy(maskpointer, texbytes, 0, texbytes.Length) returns no usable data.
    • No GPU equivalent → it’s impossible to use the mask data effectively.

PointCloud

  • No visible point cloud rendering in the [PointCloud](https://www.stereolabs.com/docs/unity/samples#point-cloud) sample.
  • ZEDPointCloudManager.cs has all its references correctly assigned (Camera, ZedManager, Material).
  • Position can logs → but nothing is rendered, in either the Built-in or URP pipeline.
    *When I use ZEDFusedPointCloudRenderer.cs to display something, the rendering is way too slow for interactive use (an issue you’ve already acknowledged but it remains unresolved).

Object Detection

  • Unity editor always crashes when entering Play Mode with the MULTI_CLASS_BOX_ACCURATE model.
  • Switching to MULTI_CLASS_BOX_MEDIUM allows Play Mode to run, but:
    • Detections are almost non-existent.
    • Enable Segmentation turned on → still no visible output.

The scripts themselves are nicely commented, but there is no clear guidance for setting up these scenes properly.

The official documentation (link) is still too superficial to solve these blocking points.

What I’d like to know

  • Are these samples supposed to work out of the box?
  • If yes → could you please share step-by-step instructions to ensure correct rendering (point cloud, object detection)?
  • If not → what is the recommended setup to get a stable and usable result today (suggested render pipeline, Unity version, earlier SDK version, upcoming hotfix, etc.)?