Isaac_ros_centerpose detection issues

[amanickapraveen]

I followed the isaac_ros_centerpose QuickStart tutorial, and was able to run the demo to detect the shoes’ pose. I wanted to replace the shoe model with other models, to detect the pose of cereal boxes, cups, and bottles, but realised the “ deployable cups dla34” model was missing. Instead I used the “Multiple 3D Centerpose” model, and it was not able to detect the pose of cups/bottles effectively. It did detect the pose of the cereal box, somewhat, but the node would crash within the docker container after about 10-15seconds, and needs to be restarted.

I do see a “centerpose_trainable_dla34.pth” file in the “Centerpose” model of the NGC. Is there any tutorial on how we can use this to train on our own custom data? Would be very helpful.

Thanks!

[Ahung]

Hi @amanickapraveen

Thank you for your post. I forward your question to the engineering. I let you know when I’ve updates.

Best,
Ahung

[Morganh]

@amanickapraveen

s there any tutorial on how we can use this to train on our own custom data?

Yes, we have notebook for it. Please see .ipynb and specs in tao_tutorials/notebooks/tao_launcher_starter_kit/centerpose at main · NVIDIA/tao_tutorials · GitHub. The doc is in CenterPose - NVIDIA Docs. The model card is in CenterPose | NVIDIA NGC.

[yingliu]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks