detect_ros.py inference time slower than detect.py

[terrytao19]

Hi, I'm running on a custom v7-tiny model, and the inference times when running detect_ros.py is at around 500ms while the inference times on detect.py is around 150ms, I added this to track the times:

[lukazso]

Hi @terrytao19, can you post the image size before feeding the image into the network in both detect.py and detect_ros.py?

[terrytao19]

Hi @lukazso , I am using the default 640 image size, I stopped running on vm so my speeds are faster but there is still a difference:

On the left is stock yolo while the right is running on ros.

[lukazso]

Please check the image size in the code. E.g. print the image size in both scripts right before model inference.

[psykaunot]

Hello @lukazso , I have the same problem, but I cannot fix it. My input images are all 640 x 480 and after rescaling, I obtained 640 x640.
This is what is obtained from the detect_ros.py script.