Hello Luxonis Community,

I hope you're all doing well. I've been exploring the capabilities of the OAK-D S2 PoE device and have found it incredibly useful in addressing various design challenges. Thank you for developing such a versatile tool.

Currently, I'm focused on implementing 3D object detection, either on the host or on the device itself. Upon reviewing the example usages provided, I noticed a strong emphasis on 2D object detection on the image plane, with subsequent depth data registration to achieve 3D detection. However, I did not come across any examples specifically demonstrating single stage 3D object detection directly on point clouds or depth maps.

Could someone please confirm if there are any such example usages available that I might have missed? Additionally, I would greatly appreciate any advice or recommendations regarding implementing single stage 3D object detection where the output is in real-world coordinates.

Thank you in advance for your help. I'm looking forward to your insights.

BR,

    Hi aytek

    aytek Could someone please confirm if there are any such example usages available that I might have missed?

    No, unfortunately there aren't any 3D detection examples to my knowledge.

    aytek Additionally, I would greatly appreciate any advice or recommendations regarding implementing single stage 3D object detection where the output is in real-world coordinates.

    This would defeat the whole point of using a stereo camera no? The real-world coordinates would probably be incredibly inaccurate.
    Generally implementing a custom model should work (as long as all layers are supported), but you would need custom decoding. And watch out for model performance while running edge.

    Thanks,
    Jaka