Can PCN-1 be used as axis aligned face detector on it's own?

[mrgloom]

I wonder can PCN-1 be used as axis aligned face detector on it's own? By looking at the output of PCN-1 even at threshold of 0.99 I can see that it gives too much false positives, I suspect it's duo to cascade nature of training, i.e. it's ok that 1st stage extract more windows, because they will be filtered by subsequent stages.

Also as I understand PCN-1 is similar to method proposed in "Multi-view Face Detection Using Deep ConvolutionalNeural Networks" paper, but in PCN-1 we also have additional bbox regression output that might help.