Course 4, Week3: question on the size of the bounding box

I don’t understand how the algorithm can detect the size of the bounding box if part of an object is outside of the grid, in which the center of the object is located (at 11:27 minute in Course 4, week3 video: “Bounding box predictions”). I understand that when a grid is passed into a conv-net, it can detect the center of the object. However, a portion of the object is outside of current grid and was located in a neighboring grid (like the image cropped from Andrew’s slide). How the total bounding box size is determined in the algorithm?

Thank you!

Sliding windows does pass an image region into the network, and it doesn’t do well with objects that weren’t fully within the region. In contrast, YOLO does not pass an image region into the network. It passes the entire image. Each grid cell is predicting at the same time on the same input.

Thanks, ai_curious. I went through several threads you and others posted. Need some time to digest them. :slight_smile: