Right, single exposure, 1/50 with 24mm/2.8 and 0.5" with 50mm/1.8. The former can adjust the aperture smoothly, while the latter can only do discrete steps (therefore the larger exposure time). Tuning has to be fine-tuned manually for each shutter speed (at least at this stage).
There was no postprocessing, other than developing the raw files (CR2 or DNG) in ufraw. The creamy bokeh images were actually full-res silent pictures with variable aperture. Unfortunately I couldn't get it to work with regular CR2 files.
The same trick is useful for getting full-res silent pics with normal exposure at hand-holdable shutter speeds, although with a pretty interesting bokeh. It simply closes the aperture to f/22 after some time, simulating a leaf shutter (one that doesn't fully close, so it still leaks some light after the exposure "ends").
Background: with current FRSP implementation, which you all know it gives overexposed images, the exposure starts at the same time on all rows (global shutter when starting the exposure), but ends at different times (gradient, top line is read out first, then subsequent lines are going to capture more light - rolling shutter). With regular pictures, the mechanical shutter determines what exactly is captured from the outside world (that's why you don't see any gradient in a regular CR2), but the sensor is "open" for a longer time. See
timing analysis for more details.
Some related fun stuff:
http://www.aggregate.org/DIT/ccc20140116.pdf
