If we know that, (in video mode), the sensor reads out a native 1620x1080 from the CMOS sensor, ideally we should be writing this stream to SDHC & not the artificially interpolated 1920x1080 file. All up-scaling would be done in software using something like a Lanczos resizing algorithm. You would then crop to a 16:9 aspect ratio.
i-frame only h.264 compression, similar to the low GOP method our GH2 cousins are currently using.
Issue #1030 on hold