use conv layer to perform regression instead of fully-connected layer?

[yijunCai]

The original YOLO use two fylly-connected layers to perform regression, while this implementation use conv layer instead. What's the point of this change? Can conv regression capture global information as fully-connected regression?

[Harick1]

@yijunCai I use convolutional layer in order to decrease the number of the parameters. And for googlenet, I think the outputs of the pool5 contain some global information in a degree, since the receptive field of pool5 is large enough.