Question about architecture

[marcvoegele-fhnw]

As I understand correct. A run-of-the-mile object detection works with conv-layers to recognize patterns. Then there is a flatten-layer and a dense layer, which connects the detected pattern to a coordinate (e.g. 300 pixel in a 256X256 pixel image is equivalent to the 2x46 coordinate.

But how does this architecture does the flatten part.
I don't understand the architecture.