How to improve Deepstack AI with 4k cameras?

[borked4days]

I recently added my first 4k cameras, previously was only using three 5mp cams. The 5mp cameras did very will with detection during the day and night. However once I added the 4k cams I've been disappointed with the detection. I don't fully understand, I would have assumed that more data would mean better detection, but instead I think more data means smaller image relative to the total image size so worse detection. I have been passing HD images to Deepstack, this massively improved my detection on my 5mp cameras, wondering if maybe I shouldn't do that with the 4k cameras. Wondering what else I might tweak to help with detection on these 4k cameras. I am running the standard object detection and deepstack dark. Not sure if I should be using other models or not.

 

[TBurt]

I recently added my first 4k cameras, previously was only using three 5mp cams. The 5mp cameras did very will with detection during the day and night. However once I added the 4k cams I've been disappointed with the detection. I don't fully understand, I would have assumed that more data would mean better detection, but instead I think more data means smaller image relative to the total image size so worse detection. I have been passing HD images to Deepstack, this massively improved my detection on my 5mp cameras, wondering if maybe I shouldn't do that with the 4k cameras. Wondering what else I might tweak to help with detection on these 4k cameras. I am running the standard object detection and deepstack dark. Not sure if I should be using other models or not.

It shouldn't matter the resolution of the camera (to a point) as Blue Iris /DS resizes all images down to 640 or less depending on your settings. (High/med/low) I would use one of Mikes models, or yolov5m.pt, or yolov5l.pt, yolov5s.pt depending on what hardware you're using.