Warn if object detection video mode (width or height) does not match model size

[mcm001]

If we feed a 640x640 model a 640x480 image we only have to pad, not also resize. This can reduce latency and overhead. Show a little warning if this is not the case