It still interferes with itself, and that interference affects the pattern of detections. It's as if the photon were a wave right up until the moment of detection, at which points it's forced to “particalize” and pick a spot to be located at — but it's the amplitude of the wave it was just before detection that determines where on the detection screen the photon is likely to show up. If you send many photons through one at a time, the detections (each just a point on the screen) will fill out the expected double slit pattern.