Hi,
I do believe, that the real collapse of the function happens in the detector. If you assume, you'd split the light in a polarizer, and then you would join the two channels again, I'm quite sure you'd see an interference pattern (u'd have to rotate the two polarisations to the same...