NVIDIA-ISAAC-ROS / NVIDIA-ISAAC-ROS/isaac_ros_pose_estimation

CenterPose Inference Shift in Isaac ROS

Open
#68 1 comment 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
501
Forks
57
PR merge metrics
No merged PRs in 30d

Description

I am experiencing an issue where the inference results of CenterPose in the Isaac ROS environment are shifted to the left and right.

model file: cereal_box_FAN_small.onnx
Image Image

model file: laptop_FAN_small.onnx
Image

When visualizing CenterPose in TAO 5.5, there was no such shift.
model file: cereal_box_FAN_small.onnx
Image

Both environments are using the same CenterPose model for inference, yet this issue only occurs in Isaac ROS.

I have also attached a video. It seems that there is no vertical shift.
https://github.com/user-attachments/assets/5a2b9dc8-6ff8-4f7d-9fb4-7322fce7d1a7

Is there a known reason for this behavior, or any recommended way to correct the shift?
Any guidance would be greatly appreciated.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the reported behavior with cereal_box_FAN_small.onnx and laptop_FAN_small.onnx in Isaac ROS, then compare the outputs with TAO 5.5 using the same models. Use the attached images and video to verify the horizontal shift and absence of vertical shift; done means the Isaac ROS inference results are aligned with the reference output or the cause is documented.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
computer-vision, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.