When using pose inference, the bounding box for the detected object is fixed at the optical center of the image. It does not overlay on top of the object. Additionally can I get some help with the following parameters in the inference json file:
“detection_pose_estimation.viewers”: {
“Detections3Viewer”: {
“frame”: “camera”,
“object_T_box_center”: [1, 0, 0, 0, 0, 0, 0],
“box_dimensions”: [6.0, 3.0, 1.8],
“mesh_name”: “green”
}
Shouldn’t the object_T_box_center be varying with each image if that’s the translation factor between object center and bounding box center?
Is it the 2D or the 3D bounding box that is fixed at the optical center of the image? And are you running the default app for dolly or boxes or a different custom object?
Regarding object_T_box_center, no that one is fixed as it is the relative pose of the defined box center with respect to object. It is usually zero translation and rotation pose unless the object center in the CAD model is not located at the actual object center in which case you have a translation offset and rotation is always zero.
It is the 3D bounding box fixed at optic center. I am running it for a custom object. I have tried giving 0 translation and rotation pose but it still does not overlay.