Hallo,
I don't understand how depth-value is computed by using 2D Bounding Box detections. I understand one may have to choose a reference point or region from each 2D Bounding Box and get the corresponding depth-value from the Depth map. But how this reference point or region is chosen? I cannot find any descriptions…
Depth-value in SpatialDetectionNetwork?
Hi mzbt
Afaik when bounding box is inferred by the detection part of the node, some smaller portion of that bbox rectangle (defined by setBoundingBoxScaleFactor()
) will be fed into the spatial part (essentially a spatial calculator node). This smaller region of the RGB image (aligned with depth map), can be taken to acquire the average Z value of the object. At the same time X and Y are computed from Z and FOV.
Thanks,
Jaka