I have previously demonstrated the dynamic range increases that come with Sony a7RIII pixel shift. A Facebook user said that there were no similar improvements with Fuji’s GFX pixel shift. That didn’t fit with what I remembered, but I hadn’t done any quantitative testing. So I thought I’d rectify that.
I made single and pixel shifted images of the back of the body cap on a GFX 100S at ISO 1000 and 1/100 second with 14-bit precision.
I used the Fuji pixel shift program to make both 4-way and 16-way DNGs. I looked at the files in RawDigger.
The single shot image has 14 bits of precision. The 4-way shifted image has 16 bits of precision. The dynamic range of the single-shot image is 15360/22.3, or 9.43 stops. The dynamic range of the red channel of the 4-way image is 61440/16.8 or 11.84 stops. The dynamic range of the green channel of the 4-way image is 61440/14.9 or 12.01 stops. The dynamic range of the blue channel of the 4-way image is 61440/17.2 or 11.80 stops.
The 16-way shifted image has 16 bits of precision. The dynamic range of the single-shot image is 15360/22.3, or 9.43 stops. The dynamic range of the all the channels of the 16-way image is 61440/101.5 or so or 9.24 stops. If we make prints of the same size from the 16-way and single-shot images, the noise in the 16-way image will be reduced by 1 stop, so, when compared to the single-shot image we should call the DR 10.24 stops.
Both pixel shifted images have better DR than their single-shot counterparts. But that’s sort of an apples to oranges comparison, since the pixel shift image is in a CIE color space, and the raw image is not.
My next move was to make a seven-stop underexposed image of my bookcase with the GFX 100S and the 30 mm GF lens at f/5.6 and ISO 1000. I made 4-way and 16-way pixel shifted versions. I brought the images into Lr, turned off sharpening and noise reduction, boosted the shadows to +100, and boosted the Exposure to +5. The white balance was different for each image, so I white balanced to about the same visual place.
Now we’ll look at some crops at slightly more than 1:1 for the single shot and 4-way images, and 1:2 for the 16-way image to keep the same filed of view:
And at about twice that magnification:
The visual effects of the noise are less apparant in the pixel shifted images.
Brian K says
I’m confused and want to clarify. The GFX100s always requires 16 exposures, correct? There’s no firmware update that changes that? However when using the software there’s a choice between keeping the file at 11648 pixels wide (I assume that’s what you mean by 4 way) and 23296 pixels wide (I assume that’s 16 way). But either result requires 16 exposures.
Rico Pfirstinger says
I remember testing this more than a year ago with PSC version 1.0 and a GFX100. I found the practical DR gain in the composite DNG compared to a single RAW pretty notable: