The CPU usage of the “traffic_light_classifier” ROS package reaches 16% when it is running.

[0116lhc]

When I was testing the “traffic_light_classifier” software package, I found that when detecting traffic lights and using the CNN detection mode, the CPU usage of this node is very high, up to 16%, and the memory usage is also relatively large. After many tests, I got the same result. I am very puzzled. Why does a CNN detection function have such a high occupancy? The detection node of YOLOX has a occupancy of less than 1%. How can this situation be solved?

[Shin-kyoto]

@0116lhc

Thank you so mush for your question.
First of all, we want to know the actual situation. Can you ask the questions below?

• Q1. Can you share the information about your computational environment?
• Q2. Can you share the rosbag to reproduce that situation? And please share the commit hash of the autoware you used.