Detect all primary objects in the image and output their 2D and estimated 3D spatial bounding coordinates.

Return a normalized JSON array with the following schema for every detected object:
[
  {
    "box_2d": [ymin, xmin, ymax, xmax],
    "label": "string description",
    "spatial_depth_estimate": "foreground | midground | background",
    "relative_orientation": "facing_camera | profile | turned_away"
  }
]
Values for box coordinates must be normalized between 0 and 1000.