CCTV System with Yolo

🤝 Community Help
1

Hello everyone,

I would like to ask for advice regarding image resolution in Roboflow. My project involves a real-time CCTV camera system with a 4K camera feed. However, when I trained my YOLO model using the default 640 × 640 resolution, the performance was not satisfactory. The model struggles to detect small objects, such as masks, and the overall accuracy is lower than expected, even for larger objects like people.

I checked the properties of my original images, and they are approximately 1594 × 894 pixels. However, Roboflow preprocessing resized them to 640 × 640 during dataset generation.

I would like to know:

  1. How can I properly change the training resolution in Roboflow (for example, to 1280 × 1280) while ensuring that I am using the original image details rather than simply enlarging the 640 × 640 images?

  2. Is training with 640 × 640 still acceptable when deploying the model on a 4K real-time camera stream?

  3. What are the recommended approaches to maintain good detection accuracy when the input camera resolution is much higher than the training resolution?

Any guidance or recommendations would be greatly appreciated. Thank you.

2

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.