using image segmentation result instead of object detection

[lturing]

according to "The pixels within the same 2D bounding box are assigned with the same depth label of the corresponding object", so if using image segmentation, the result will better?