Bad results with custom data

[ronnypollak]

Hello, I am currently trying to develop a custom workflow which includes the pose estimation as the final step.

I’ve successfully implemented this process before with a different object using the same steps, but I’m encountering issues with my current object.

I wanted to ask if anyone can help me with my current problems.
The pose estimation is not working correctly.
I suspect that the depth images might be the source of the problem but I am unsure if that is the case and if so what the reason for it is.

My debug folder:
https://drive.google.com/drive/folders/1zrUQGApq5PrzUHH52fHefH_xU4zc7cpo?usp=drive_link

Example of the depth images I used:
https://drive.google.com/drive/folders/1SHk_EcXYHpT-Yz-UxmjMtI4YfBa-I1vX?usp=drive_link

In my setup I am using ARCore to predict the depth images. With my first test object this process worked perfectly fine, so now I don't know what the problem is.

[wenbowen123]

Hi, your depth image does not look right. You can visualize the file scene_complete.ply to tell. You can check the troubleshooting for more details.

[lexuanhoang04]

Hello, I am currently trying to develop a custom workflow which includes the pose estimation as the final step.

I’ve successfully implemented this process before with a different object using the same steps, but I’m encountering issues with my current object.

I wanted to ask if anyone can help me with my current problems. The pose estimation is not working correctly. I suspect that the depth images might be the source of the problem but I am unsure if that is the case and if so what the reason for it is.

My debug folder: https://drive.google.com/drive/folders/1zrUQGApq5PrzUHH52fHefH_xU4zc7cpo?usp=drive_link

Example of the depth images I used: https://drive.google.com/drive/folders/1SHk_EcXYHpT-Yz-UxmjMtI4YfBa-I1vX?usp=drive_link

In my setup I am using ARCore to predict the depth images. With my first test object this process worked perfectly fine, so now I don't know what the problem is.

Hi, I want to ask for your pose estimation of your custom dataset, are you using the model-based or model-free Foundation Pose? If you are using model-free Foundation Pose, how do you prepare cam_in_ob? Moreover, I don't know if this is true but I think cam_in_ob is already the pose