X-Trans vs Bayer

yogi4fitness

Veteran Member
Messages
2,522
Reaction score
1,561
Location
MOUNTAIN VIEW, CA, US
I keep reading about how Fuji X Trans is better than Bayer CMOS for sharpness and color rendition. Is that really true or simply marketing speak? If it really is superior, will there be a medium format X Trans sensor in the future for the GFX line?
 
I keep reading about how Fuji X Trans is better than Bayer CMOS for sharpness and color rendition. Is that really true or simply marketing speak? If it really is superior, will there be a medium format X Trans sensor in the future for the GFX line?
No, it is not true.

Both Bayer and X-trans are compromises. The advantage of X-trans is that reduces the risk of color moiré.

The downside is that some colors are undersampled, red if I recall correctly and it needs different interpolation than Bayer.

Fujifilm says that MFD does not need X-trans, but there is probably some marketing statement in that.

Best regards

Erik
 
I keep reading about how Fuji X Trans is better than Bayer CMOS for sharpness and color rendition. Is that really true or simply marketing speak? If it really is superior, will there be a medium format X Trans sensor in the future for the GFX line?
No, it is not true.

Both Bayer and X-trans are compromises. The advantage of X-trans is that reduces the risk of color moiré.

The downside is that some colors are undersampled, red if I recall correctly and it needs different interpolation than Bayer.

Fujifilm says that MFD does not need X-trans, but there is probably some marketing statement in that.

Best regards

Erik
Yeah, the real answer is probably "Our engineers saw some benefits to using Xtrans, but it is a way to differentiate ourselves in the market and we weren't able to get a custom chip for the lower volume MF sensors, so Bayer works just fine."

I've shot Xtans before and had no problems with it except for finding high quality demosaicing in my workflow. The Xtrans helps with some types of aliasing, hurts with others.
 
I keep reading about how Fuji X Trans is better than Bayer CMOS for sharpness and color rendition. Is that really true or simply marketing speak? If it really is superior, will there be a medium format X Trans sensor in the future for the GFX line?
No, it is not true.

Both Bayer and X-trans are compromises. The advantage of X-trans is that reduces the risk of color moiré.

The downside is that some colors are undersampled, red if I recall correctly and it needs different interpolation than Bayer.

Fujifilm says that MFD does not need X-trans, but there is probably some marketing statement in that.

Best regards

Erik
Yeah, the real answer is probably "Our engineers saw some benefits to using Xtrans, but it is a way to differentiate ourselves in the market and we weren't able to get a custom chip for the lower volume MF sensors, so Bayer works just fine."

I've shot Xtans before and had no problems with it except for finding high quality demosaicing in my workflow. The Xtrans helps with some types of aliasing, hurts with others.
Is Xtrans more efficient? Is that the reason the base ISO is 200?
 
I never liked the rendering of the x-trans sensor, I really hope Fuji stick to the bayer sensors in there top line products. The sensor in my GFX100S is superb.
 
Yeah, the real answer is probably "Our engineers saw some benefits to using Xtrans, but it is a way to differentiate ourselves in the market and we weren't able to get a custom chip for the lower volume MF sensors, so Bayer works just fine."

I've shot Xtans before and had no problems with it except for finding high quality demosaicing in my workflow. The Xtrans helps with some types of aliasing, hurts with others.
Is Xtrans more efficient? Is that the reason the base ISO is 200?
Shouldn't be. Xtrans has slightly more green pixels but that shouldn't make a big difference. The X-T4, which has the same pixel architecture as the GFX 100 has base ISO of 160. I don't know why.
 
I have to say that I think Erik has written a very short and nicely done little reply here on XTrans. I stared at it a while and looked at my long and definitive list (compiled over 8 years of intensive research and DPR-tracking 🕵️‍♀️) of XTrans advantages and disadvantages, and he perhaps missed some of the "strengths" but then again, I have come to believe over the years that XTrans does have a significant marketing element to it, so he is right about that.

I'm not bragging, but I'm just telling you that no one on this orb we call Planet Earth shot Fuji X more than me from 2013 up until two years ago when I stopped shooting Fuji X gear. I shot it like a madman all over the World and I had all of the gear.

I was a big believer in and supporter of the XTrans thing because I liked the rendering and thought it gave the smaller sensor (compared to FF) an ability to get close to (or even sometimes match) the image fidelity of FF while enjoying the advantages of the smaller Fuji X system and great XF glass.

I believed that the people who attacked XTrans were almost always really just conducting marketing attacks on Fuji itself and that they had even less actual technical prowess than even me (which is a pretty low bar). I got in a lot of trouble over the years fighting with anti-XTrans (anti-Fuji marketing dudes) on the Fuji Board. This was mostly from 2014 through about 2019 - a 5 or 6 year period where I defended XTrans and considered myself the World's top authority on XTrans (😲), even though I had absolutely no idea how it worked technically. I simply had a list of every positive aspect ever mentioned on the DPR Fuji Board, Fuji reviews throughout the web and statements of Fuji themselves, as well as the rare actual technical article (which were all over the map and often depending on the bias of the author). I also had the list of all the attack-points people used against X-Trans. I tossed that around and used them (and my own shooting experience with Fuji X gear) to try to combat the anti-XTrans marketing frenzy that occurred on the Fuji Board on a constant basis for 6 years (and no doubt continues today - but I don't dwell there anymore so could be wrong).

I don't care anymore, but here are a few of my basic observations after thinking about it over a long period of time (years) and shooting a lot of Fuji X, Canon DSLR FF (5D II, III and IV) and now Q2 and GFX.

I think XTrans is good and helps stretch the capabilities of that smaller sensor to get close to FF image fidelity in a lot of situations. I compared thousands of FF Canon DSLR (5D) shots (not 50 MP) to Fuji APSC X-Trans shots over the years and thought the Fuji X shots held up very well to the Canon FF shots in a variety of situations. That was just me peeping my own shots and was not a scientific comparison of course. Anyway, I liked XTrans.

Now that said, there was a huge disadvantage of XTrans and that was Adobe in the early years of Fuji X. Lightroom had a terrible time rendering XTrans files and it resulted in the famous watercolor effect on foliage and brick in some situations. That got better over the years and I don't think was an issue the last year that I shot Fuji X.

In my opinion, about 85% of what has been written about XTrans on the DPR Fuji Board is nonsense and just a bunch of pro-Fuji / anti-Fuji marketing speak. But I haven't followed it in the past year or so.

I think XTrans enables Fuji X to punch above it's weight class. It is good. Adobe has basically )over time) negated the old demosaicing issue and I like the rendering of XTrans files in most situations. I think the image fidelity of Fuji X is pretty remarkable for that smaller sensor and the Fuji one-two punch of X-GFX is quite a brilliant thing and hits the FF market from both flanks like Hannibal at the Battle of Cannae.

One last point.... No way Fuji ever even briefly considered XTrans for GFX. I'm glad Fuji uses it for their X line and not for GFX. Most people I know that shoot both systems agree with that sentiment.

--
Greg Johnson, San Antonio, Texas
https://www.flickr.com/photos/139148982@N02/albums
 
Last edited:
X-Trans is a baby step.

Here's a CFA designed to have minimal color moire:

0f2bcdfe003c46dfa9e87dcdb3805068.jpg.png

Close view of a crop:

4e8d7f0b1f6d4234876b48eb81992bfd.jpg.png



If you want to generate your own CFAs, here's the code:



83e5068fe6774166890209cdc78ac792.jpg.png





--
https://blog.kasson.com
 
Last edited:
Hi,

I had a Drag Car at one point which was really into making ExTrans....

Catastrophic Self Disassembly.....

Sorry. This is what pops into my head when this subject comes up....

Stan
 
Hi,

We can always go back to monochrome sensors with color wheels. And hope nothing moves any at all....

I had a Kodak 460m with a wheel at one point. I think that was the last model which worked that way. I added a 460c and took the wheel off the M and used it normally for B+W.

Stan
 
Hi,

I had a Drag Car at one point which was really into making ExTrans....

Catastrophic Self Disassembly.....

Sorry. This is what pops into my head when this subject comes up....

Stan
Stan - did you race midgets, too?
 
Hi,

Not me. That was Dad. And Uncle Bob. 1948 until....well, not sure. Then there were Stock Cars until 1964. I began in 1982, and that was sports and formula cars to 1994. Then two years away from it while I moved from NY to NC. Then I got into Legends Cars quite a bit but still some other things. Including a short lived Drag Car. The most costly car to operate per mile ever. -boom-

I pulled off the track after the 2005 season.

Anyway, y'all watch those Googly Bear searches. You'll get my dad as often as you'll get me.

Stan

--
Amateur Photographer
Professional Electronics Development Engineer
Once you start down the DSLR path, forever will it dominate your destiny! Consume
your bank account, it will! Like mine, it did! :)
 
Last edited:
Hi,

We can always go back to monochrome sensors with color wheels. And hope nothing moves any at all....

I had a Kodak 460m with a wheel at one point. I think that was the last model which worked that way. I added a 460c and took the wheel off the M and used it normally for B+W.

Stan
 
Eh, the code to 'design' the sensor filter is the trivial part. You need to post the code to demosaic such a sensor--yeah, I realize it won't fit into a post here, maybe try GitHub?--and some of your famous test results showing that your code produces a credible result (simulated is okay, I trust you).

My guess is that the "your code produces a credible result" part, relative to real-world photos and/or test charts, is a Ph.D. dissertation-level effort. But what do I know?

Anyway, keep thinking outside the box, Jim!
 
Eh, the code to 'design' the sensor filter is the trivial part. You need to post the code to demosaic such a sensor--yeah, I realize it won't fit into a post here, maybe try GitHub?--and some of your famous test results showing that your code produces a credible result (simulated is okay, I trust you).

My guess is that the "your code produces a credible result" part, relative to real-world photos and/or test charts, is a Ph.D. dissertation-level effort. But what do I know?

Anyway, keep thinking outside the box, Jim!
There is a paper on how one can evaluate new CFAs...


But it is interesting that the Bayer Array was developed in the 70s and no new CFA has been able to replace it in all that time. Sure there have been some 4 color arrays, X-Trans, etc but nothing has gotten the wide spread acceptance that the 3 color Bayer array has enjoyed.

Maybe because one needs wide spread software support to launch a new CFA ...
 
You need to post the code to demosaic such a sensor--yeah, I realize it won't fit into a post here, maybe try GitHub?--and some of your famous test results showing that your code produces a credible result (simulated is okay, I trust you).

My guess is that the "your code produces a credible result" part, relative to real-world photos and/or test charts, is a Ph.D. dissertation-level effort. But what do I know?
There is a paper on how one can evaluate new CFAs...

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0232583

But it is interesting that the Bayer Array was developed in the 70s and no new CFA has been able to replace it in all that time. Sure there have been some 4 color arrays, X-Trans, etc but nothing has gotten the wide spread acceptance that the 3 color Bayer array has enjoyed.

Maybe because one needs wide spread software support to launch a new CFA ...
Thanks, I may check that out. But to my simple mind, the CFA and the demosaic algorithm can only be evaluated as a functioning pair. That's not to say better demosaic algorithms may not, in time, be devised--just that whether to adopt a CFA design in the first place seems to me highly dependent on how well it can be demosaiced with an existing algorithm.

As for the endurance of the Bayer CFA, I suspect part is that it's basically a sound and simple design; and part is that its ubiquity drives development of demosaicing algorithms and software generally, in a self-reinforcing situation.

But again, those are just my personal suspicions--I claim no expertise!
 
Last edited:
Eh, the code to 'design' the sensor filter is the trivial part. You need to post the code to demosaic such a sensor--yeah, I realize it won't fit into a post here, maybe try GitHub?--and some of your famous test results showing that your code produces a credible result (simulated is okay, I trust you).
Uh, this was a joke.

A bad one, if I have to explain.
My guess is that the "your code produces a credible result" part, relative to real-world photos and/or test charts, is a Ph.D. dissertation-level effort. But what do I know?

Anyway, keep thinking outside the box, Jim!
 
Eh, the code to 'design' the sensor filter is the trivial part. You need to post the code to demosaic such a sensor--yeah, I realize it won't fit into a post here, maybe try GitHub?--and some of your famous test results showing that your code produces a credible result (simulated is okay, I trust you).
Uh, this was a joke.

A bad one, if I have to explain.
Story of my life.
 
Eh, the code to 'design' the sensor filter is the trivial part. You need to post the code to demosaic such a sensor--yeah, I realize it won't fit into a post here, maybe try GitHub?--and some of your famous test results showing that your code produces a credible result (simulated is okay, I trust you).
Uh, this was a joke.

A bad one, if I have to explain.
Sorry, sailed over my head. I thought that you / someone had a serious interest in very complex or even totally-random CFA patterns. I can imagine some real advantages, albeit at the cost of exponentially more complex demosaic algorithms for a given level of image quality.
 

Keyboard shortcuts

Back
Top