Isaac_ros fopundationpose detection speed FPS is lower than the performance results

Hi, when I test in the RTX 4090, I found that the speed of the pose detection of foundationpose is only reach the FPS 1.4-1.6, but in the document of isaac_ros the speed can reach the 7.02 PFS , I cant find the reason for that,below is my result of the test. Thanks for attention to my question.

Hi @Miahhhh

Thank you for writing this post.

First question: have you tried our benchmarks to see if your setup works?

You can find here how to set your benchmark: Benchmarking — isaac_ros_docs documentation

Another reason may relate to other software running simultaneously; please investigate using tools like:

  • htop
  • nvtop

Let me know

Best,
Raffaello

Hi @Raffaello I use Benchmark to test foundationpose in my hardware,the reason as follow ,it’s 5.7FPS nearly to the official reason 7FPS. but I still confuse for that I run the foundationpose example

ros2 launch isaac_ros_examples isaac_ros_examples.launch.py launch_fragments:=realsense_mono_rect_depth,foundationpose mesh_file_path:=${ISAAC_ROS_WS}/isaac_ros_assets/isaac_ros_foundationpose/Mac_and_cheese_0_1/Mac_and_cheese_0_1.obj texture_path:=${ISAAC_ROS_WS}/isaac_ros_assets/isaac_ros_foundationpose/Mac_and_cheese_0_1/materials/textures/baked_mesh_tex0.png score_engine_file_path:=${ISAAC_ROS_WS}/isaac_ros_assets/models/foundationpose/score_trt_engine.plan refine_engine_file_path:=${ISAAC_ROS_WS}/isaac_ros_assets/models/foundationpose/refine_trt_engine.plan rt_detr_engine_file_path:=${ISAAC_ROS_WS}/isaac_ros_assets/models/synthetica_detr/sdetr_grasp.plan the speed is only 1.5FPS . I use nvtop, and the GPU utilization is only about 15%. Could you help me to analysis the reason?


The reason can be related to other software running simultaneously and using the GPU.
Meanwhile, I will forward your issue internally if there are other related reasons.

Raffaello

But I use he nvtop to monitor the GPU, and the other software have very low utilization, and the GPU is not fully occupied.