I hope you are all doing well. I’m reaching out to this community to seek your advice and insights regarding a project I’m currently working on.

Thank you in advance for any help or suggestions you can provide!

I have an ensemble of three computer vision neural networks, and I’m working on connecting over 500 cameras. Currently, I’m testing with 5 cameras, using the OpenCV library to receive RTSP streams. I then load the neural networks using the Ultralytics library and process each frame through them. The results are drawn as bounding boxes with OpenCV, and I display the images using Streamlit.I’m testing this setup on a simple PC with a 12 GB GPU, while waiting for a server with two GPUs totaling 98 GB. However, the performance is very slow, and at times, the system completely freezes.

As a beginner in computer vision, could you please recommend libraries or approaches to improve performance, considering the future load of 500 cameras?I plan to switch from OpenCV to a GPU-supported library. Would exporting models to ONNX and running them through ONNX Runtime help with performance?Has anyone organized a similar system? Any advice would be greatly appreciated!