ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question

Why tensorflow python3 script runs slower in ROS than standalone?

asked 2019-07-29 12:16:21 -0500

npscars gravatar image

updated 2019-07-29 17:02:22 -0500

Ubuntu 18.04 ROS Melodic Python 3.6.8 Tensorflow: 1.13.1 TensorRT: 5.1.5 NVIDIA: GTX1060 using Cudnn 7.6 and Cuda 10.0

When I run a object detection model developed in python using tensorflow framework in ros, the speed appears to be slower than if i run the same tensorflow model in pure python environment. Is this expected behaviour?

I tried tensorflow object detection model named "ssd_mobilenet_v2_coco_2018_03_29" python3 --> takes 16ms approx. rosrun inference_node --> takes 22ms approx

Just for information my catkin workspace is build with python3 too.

For example my model stats below python3 ---> takes 60ms approximately But rosrun inference_node and subscribing to an image broadcasted by a bag file --> takes 85ms approx.

This seems strange or I am missing some point.

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted

answered 2019-10-20 11:53:44 -0500

npscars gravatar image

I found out the reason why this was happening was because the image size input to python3 directly was a bit smaller than the input to ros node. Sorry for confusion but thought would be still useful for someone who might experience this issue like me.

edit flag offensive delete link more

Question Tools


Asked: 2019-07-29 12:14:02 -0500

Seen: 294 times

Last updated: Oct 20 '19