I'm running object detection with coral tpu using 4K cameras. Since images have to be resized detection is quite bad.
Would it be possible in the monitor definition inside objectconfig.ini or mlapiconfig.ini to define coordinates for object detection?
It would be cool if one could define a single area in the image x1,y1,x2,y2 (or x1,y1,w,h) where the image should be cropped before it is sent to the object detection, then run detection and then project any bounding box back into the original image.
I don't think it is necessary to be able to specifiy multiple regions as you can always run detection on the whole image.
If the option is unset just use the default method.