The Bayer vs X-Trans test is carried out with a “Markesteijn algorithm”. How come we have never heard of this? Maybe because this very unknown algorithm might excel at some point (sharpness?) but performs below average on several other aspects (artifacts, etc).
Unfortunately it was not possible to do testing with the proprietary Fuji demosaicing algorithm, as explained in that article. There is some evidence provided (see his link) that the one used is no worse, and perhaps better, than the Fuji algorithm. (I would have liked to see a comparison with Iridient myself, but he gives in a another linked article a comparison of the Markesteijn algorithm with Iridient Developer, and they are virtually equivalent.) Note also that he used a suboptimal Bayer algorithm due to the limitations of DCRaw, so if anything the test contains some bias towards the X-Trans array.
For completeness I need to say that I read a lot of nonsense in the article such as “noise reduction baked in to the X-Trans demosaicking algorithm”. Noise reduction has absolutely nothing to do with demosaicking.
In the context of that paragraph, it is clear he is talking about the entire Fujifilm in-camera image processing system.
Another thing I read is “false color reduction of FujiFilm’s cameras comes not from the choice of sensor CFA”. False color can ONLY come from the CFA, without the CFA the sensor is monochrome and blind for any color info. There is lot’s of more BS.
Again, he is clearly discussing in-camera processing.
In the Iridient example I clearly see resolution differences where the X-Trans appear to be much sharper, when viewing at 100%. Also much better color separation. A find the image a bit oversharpened though, this should not have been necessary without still keeping sufficient detail.
I believe both were converted using the default Iridient Developer settings. My suspicion is that it's a demosaicing issue, not a sharpening issue per se. In any case, if even the best raw developer on the market can't get foliage quite right out of the box, as a matter of practical workflow that is not too encouraging. (Especially given the developer himself admits there are inherent limitations in his ability to demosaic X-Trans files, cf. the comment about libraries earlier.)
I have much more confidence in the tests carried out by DPreview. Have a look at:
https://www.dpreview.com/reviews/fujifilm-x-t3/7
I see a lot of Adobe samples and then a C1 noise test. We have already established that LR/ACR should not be used with X-trans files, so the Adobe samples are uninformative. And noise is not really my concern. So I am not sure what this link is supposed to show.
Let’s get back to some FACTS:
Fact-1: With X-trans the green pixels are positioned near each other. In theory this will give twice the spatial resolution with respect to a Bayer sensor.
Fact-2: The green channel contributes 59% to the luminance channel (Y= 0.30R + 0.59G+ 0.11B).
My conclusion: what comes out of the camera is anyway an image with an increased resolution.
Then it’s up to the RAW converter to utilize this. We all know that previous versions of Adobe RAW Converter were underperforming in this field. I believe this has been solved in the latest releases.
Additionally CaptureOne is doing a great job in getting out all details of Fuji’s RAF file while at the same time keeping aliasing under control.
In the end I do think that Fuji’s X-Trans CFA is a very clever design. For completeness I need to mention that with increasing pixel quantity the additional benefits will get smaller over time, not in the last place because the lens MTF will become the limiting factor, behaving as an optical low pass filter.
I agree the X-Trans array is a nice idea. Unfortunately I have not seen these theoretical benefits borne out in practice. And there are very real practical downsides, including not being able to work in LR if optimal image quality is the goal, or if one wishes to disable to the built-in lens corrections.
(I think these *can* actually be removed with some clever software, but I would need to do more work to confirm this.)
That's pretty well it. There's tons of reviews out there, I don't see many saying don't buy it.
Strangely, none of them really talk about the raw conversion issue in depth, which makes it hard to have faith in them. This seems just as important as the sensor and available lenses from the perspective of image quality.
Also note...this is highly based on your subject matter. There are plenty of Fuji shooters using Lightroom with no issue. The artifacts complained about, the most offensive ones are only on specific kinds of patterns usually. In fact, most of the time on most general subjects the files look great. And a lot of this comes down to pixel peeping. You most likely will see none of this in print or outside of 100% mag. I would rent one and see if it works for you or not.
Yeah, I am leaning this direction. It may be a compromise I can live with (just like every affordable lens ever...).