If we choose the grid box in YOLO too small , will it be worse to detect a object?

In the pic, I split the pic into a 3 by 3 grid, I knew it will be better to choose a 19 by 19.
But will it be worse to detect an object if I choose a too-dense grid, like 3000 by 3000? (Don't concern about the running time, just the accuracy of detection)

I think it will because the conv net will be spatial correlative, which means it doesn't know what happens even in the adjacent box, so it will be hard to train to get the feature of an object if too much concreting on the small detail in 3000 by 3000 boxes.

Am I right? If I make a mistake, please help me correct it, thanks!!!

Splitting an image to 3000*3000 patches for objects detection or any other tasks can affect the accuracy negatively as the information or features may be lost at the edges of the patches which will lead to the failure of capturing the details of the objects in the image, resulting in a loss of accuracy. Take for instance yoloV3 it makes predictions at three layers: the 13 x 13 layer is responsible for detecting large objects, whereas the 52 x 52 layer detects the smaller objects, with the 26 x 26 layer detecting medium objects. For more details check out this blog post: https://towardsdatascience.com/yolo-v3-object-detection-53fb7d3bfe6b. The question is why the need to split it to 3000 by 3000?