NVIDIA-ISAAC-ROS / isaac_ros_pose_estimation

Deep learned, NVIDIA-accelerated 3D object pose estimation
https://developer.nvidia.com/isaac-ros-gems
Apache License 2.0
142 stars 21 forks source link

Way to generate category labels for CenterPose ? #15

Closed ArghyaChatterjee closed 1 year ago

ArghyaChatterjee commented 1 year ago

Hello,

Thanks for your work. Is there a way to generate category labels in CenterPose as you do for DOPE during instance segmentation ? As there can be multiple categories, it's confusing if the network is not generating any labels for each category.

Thanks in advance.

jaiveersinghNV commented 1 year ago

The CenterPose model architecture requires each individual model to be trained on a specific category of object. As a result, the Isaac ROS CenterPose Node doesn't provide any category labels for its outputs; any output by the network is necessarily an object of the class that this particular network was trained on.

For example, when running the CenterPose Node following our tutorial here, the node will only output poses of shoes, since the default model used is the shoe_resnet_140.pth.

If you wish to run multiple instances of CenterPose node, each using a different model to detect a different category of objects, then you can use the ROS 2 topic remapping functionality to separate the topic outputs at a graph level.