Bounding box rotation and size is wrong for some custom objects

[monajalal]

Dear @wenbowen123

While the produced projected 3D bounding boxes are correct for some objects (e.g. blue cup), it is not correct for this custom object we had in the lab. Could you please guide what could be wrong and some guidance on how to fix it? The orientation is shown correctly.

This is actually the first frame in the capture. All of the other frames are accordingly have wrong 3D bounding boxes.

[monajalal]

I did a capture for a much simpler object and still have the same problem

In this special case, the center of object where orientation visualization is shown is also not located properly

[monajalal]

I changed the z-far from 1 meter to 1.2 and I get the bbox in correct location but not correct orientation, also when changing it to .9 caused errors. My object was approximately 1m away from the non-moving camera on a tripod.

Please let me know if you have any suggestions as how to fix the orientation of the bbox? This is btw the first frame.

I am not sure why the orientation on the first frame is wrong.

Please note that the produced mesh is correct but the bounding box and orientations are not

[wenbowen123]

the drawing of the bounding box will only be correct after you finish the entire run, since It needs the complete shape to find the correct orientation.

[monajalal]

@wenbowen123 the drawings I have shown you are done after the entire run of the first step.

[monajalal]

So, I ran the whole thing again, the mesh is not fully recovered correctly (back of mesh is weird), and the orientation (and somewhat size of) bounding box is wrong in the first frame. Could you suggest what can be fixed?

here is the link for the log of first step https://drive.google.com/file/d/1JlhCFFRuHNLWlBSC-j_rp1h_ERDaElCg/view?usp=sharing

[monajalal]

I ran the global refinement step and that didn't really refine the mesh. Which still results in the bounding box being bigger and perhaps the orientation to be wrong. Below is the mesh produced after global refinement step

After fully running the global refinement step, here is the pose viz of first frame

here is the link to log of global refinement step

https://drive.google.com/file/d/1_DneE8-fH1ljdHKV-10kl76AgsnbVpk4/view?usp=sharing

[monajalal]

I changed the z-far to 1.1 in run_custom.py (as you see from my messages above z-far being 1.1 or 1.2 has similar results)

I have this mesh which is correct for the most part

however, the bounding box is not as tight as you would expect it and its rotation is slightly off. I am showing the bbox for first frame

here's the log of step 1 https://drive.google.com/file/d/1lPlltKQ2edBd6Ymy8sJR5tTFfqqqftsA/view?usp=sharing

here's the log of global refinement https://drive.google.com/file/d/151iQgklybfxkLplaLnnZWAHQIxhavy3N/view?usp=sharing