Shared publicly  - 
 
Best Example of Fuji X-Trans "False Detail" I've Found Yet

Take a look at this image or click the link below and take a look at the studio comparison tool from dpreview for yourself. Try to ignore DPReview's tool's stitching errors, they are not relevant to this comparison.

http://tinyurl.com/falsedetail

Allow me to explain. This is a full resolution comparison of four different cameras shot at base ISO. These are out of camera JPEGs to give the best possible results for the Fuji X-E2. We have the 36mp A7r, 24mp A7, 16mp Pentax K-500 and 16mp Fuji X-E2. The selection of the image shown is a monochromatic resolution demonstration sheet, a part of the varied studio comparison scene from DPReview. What it is isn't all that important, what matters is that it is A) not colored green and B) depicts fine straight lines set at an angle in the frame.

The reason "A" matters is that finely detailed green materials look terrible on the X-Trans sensor in just about every situation (don't believe me? Select the green tufts of fake fur on the studio scene and just have look for yourself. Awful.) So here we have the best chance for the X-E2 to show more detail due to eliminating the OLPF (AA filter). The reason "B" is important is that the nature of digital images makes angled lines hard to draw sharply, you get "steps" from aliasing. The only way to avoid steps is to blur edges.

So what do we see? Well we can see very clear straight lines on the higher resolution, standard bayer array cameras. That's a big bonus of higher resolution sensors, even with blurred edges in the digital image, there is quite a lot of contrast between the lines. But, even when we look at the 16mp Pentax camera image, you still see rather straight lines though they are all gray-ish (we're looking at the finest lines here) due to the anti-aliasing effect. So the result here is more of a texture at normal viewing sizes, but, it's more accurate than the Fuji. Just look at that awful mess. At first, you might think, "Hey, look at that crisp detail on the fuji!" but, then you realize it hasn't rendered the actual detail at all. Instead, it's created an angled checkerboard that, even at smaller viewing sizes, produces an entirely different texture in the image, and of course, at full pixel resolution, is just obviously wrong.

This is a fantastic example of what happens when you take a 16mp sensor and chop it up in 3x3 sections instead of the usual 2x2, then try to cover up that massive reduction in resolution with overzealous interpolation.

The fact is, the X-trans sensor design utterly fails at its implied capability: It does not produce better details than a 16mp sensor with an anti-aliasing filter. It just produces wrong details. And this is best case. If you were to take this same image except use green lines, it would just be a blurry smudgefest due to the blocks of 2x2 green pixels on the X-Trans "semi-random" color filter array. And to add insult to injury, this sensor design does not entirely prevent moiré, which was the entire impetus behind a 3x3 semi-random color filter array.

The one benefit the X-trans sensor does seem to have is a higher base ISO. Even though the base ISO is fudged pretty hard (it's not really a full stop better than its peers, more like 2/3rds), it does give it best-in-class noise performance at high ISOs. Though there is one additional little evil detail regarding that. The process for dealing with the semi-random color filter array on the X-trans sensor forces noise reduction above the norm. However, in practice, this isn't really important. You've already lost the fine detail you might recover by using lower NR strengths anyway. And, in fact, at higher ISOs the practical impact of the detail loss is far less relevant when comparing to the technology's peers. This is why some early reviewers noted that the Fuji sensor seemed to improve as ISO's rose, but, only in comparison to peers.

Of course, this explains the demonstration that Fuji's lowest end X series mirrorless camera actually produces much better detail than the higher end models as it doesn't have an X-trans sensor. It would also explain why Fuji will produce a full frame mirrorless camera. Just look at the Soy Alpha 7 image here, see how much more vastly detailed it is. Pump up the ISO and you'll see it's on par or improved over the X-E2 despite the X-E2's NR and base ISO advantages. And remember that Fuji's "high end" X-Pro1 sold for $1700 when it launched, which is the US MSRP of the Sony Alpha 7.

Basically, Fuji has no choice if they want to maintain their position as a competitive high end mirrorless camera manufacturer. X-Trans just isn't going to cut it like this. Fuji could consider something like a 36mp sensor with X-trans to compete with 24mp full frame cameras. It would have similar effective resolution (especially if they ditch the overzealous interpolation in the JPEG engine) and possibly cleaner high ISO results.

Either way, X-trans has proven to be less than fantastic at anything but very high ISO images and really, that capability is nowhere nearly as important as so many photo blogs would have you think, especially if you can get some proper fast glass in front of the sensor.
5
3
Darrell Harris's profile photoTim Bray's profile photoCorwin Black's profile photoWill Beckley's profile photo
29 comments
 
So, start by switching to RAW, then insert a better Pentax, say the K-3. And then move around the scene, looking at objects other than slanted parallel black-on-white lines. This reveals that (a) the Sonys have a whole lot more pixels, so they have way more pixel-peeping detail (b) The Fuji colors are noticeably less saturated (c) The DPR comparo tool undercompensates for the pixels-per-inch, so the images with more pixels are irritatingly bigger in a way that makes comparison really hard (d) The Fuji stomps the others into rubble at ISO6400 (e) The A7 IQ appeals to me more than the A7R.  What I don’t see is anything objectionable with the Fuji’s treatment of the green fake-plants.  I’ve had excellent results shooting high-detail botanicals with the X-E1, so I would have been surprised if the test shots backed your claims, but they don’t.

On the other hand, if slightly-angled fine black lines at 100% are what matter to you, those Sonys are hard to beat.  
Ugo Cei
 
Commenting here so I can keep track of further comments. I presently have an X100s and an X-E2 should be delivered today, but I haven't done any serious pixel peeping.
 
"On the other hand, if slightly-angled fine black lines at 100% are what matter to you, those Sonys are hard to beat." +1
 
Interesting article and premise.  I might suggest that one item left out by DPReview and your theory is the photocell size on the sensor.  Keep in mind that larger mega-pixel sensors do not mean more detail.  However, smaller photocells mean smaller detail can be resolved.  As an example, think of a checkerboard sized photocell resolving a line from a pencil.  There will be large blocks, but no real line to see.  However, smaller blocks like graph paper might show a line at much better resolution than a checkerboard and 8 1/2 x 11 inches is smaller than a standard sized checkerboard.  Now carry this to a sensor size.  Full-frame, APS-C and 4/3 size sensors can have the same photocell size and would get the same diagonal line resolution.

However, everything comes at a price.  The smaller the photocell, the less light is captured and the more noise is present at higher ISO's.  So while extremely large photocells could have no noise present in an ISO 12,800 image, the resolution would be substandard.  It's a tradeoff.

I'd like to see comparisons of similar photocell sized camera sensors with the same diagonal line test at varying ISO's (100, 200, 800, 1600, 6400, 12,800).  Then we could see the processor at work and determine how the X-Trans sensor stacks up against equivalent sensors.  Nice work!

http://blog.outdoorimagesfineart.com
 
+David Knoble the relevance of what you're getting into is lessened by the fact that the comparison images are projected at a 1:1 object:image ratio. In other words, all of the same information is projected (as best as whatever lens is used can do) onto each sensor such that all of the scene is evenly divided across all of the pixels. Thusly, assuming lens quality is not an issue (which is not entirely safe, obviously), the division of the projected image will be equal for sensors of equal pixel count. There are some caveats, most sensors do not use all of their pixels, and as well as considering actual pixel sizes you have to consider pixel gaps and microlens designs as well.

There are definitely cases where higher density sensors appear to render more detail, such as the 24mp APS-C sensors behind a very good lens. In reality, this is more likely related to the size and design of the microlenses than any other factor. That said, Sony has introduced a gapless micro lens design for the A7r's 36mp sensor and under the absolute ideal conditions it does appear to produce a slightly smoother rendering than the sensor in the Nikon D800E (which is a good thing).

Though, in reality, the greater benefit is more total light gathering than real detail improvements in practical photography. That additional light gathering is likely overshadowed by the additional noise introduced from running the sensor constantly in a mirrorless camera which adds heat that generates noise in the image.

And for the record, the photocell size of the Fuji 16mp X-Trans, the K-500, and the A7r are all almost identical with the A7r's pixels only very slightly larger (about a 15.2mp density in the APS-C crop vs. 16mp). So it's really only the A7 that is significantly different in this comparison.

The actual effect of larger sensor sites in an image is very simple, it produces a very small reduction of effective aperture (which is why bokeh looks better and transitions more smoothly on higher density sensors) and imposes a modulation function on the image projection that is effectively the same as a simplistic sharpening algorithm.

As far as noise is concerned, once you've normalized your source data from different sensors into the same common output size, say an 8x12 print, the difference in noise due to pixel size is far outweighed by the added detail provided by more pixels which ultimately gives a much better effective noise ratio, color rendering, and dynamic range in aggregate. This is why Nikon D800/E tops the charts on DxOMark for overall scores, as all of those factors matter and the added resolution boosts color and DR to compensate for slightly lower pure SnR results.

Anyway, none of that is really relevant to what X-trans is doing to lead to the demonstrated results. The simple fact is, Fuji is dividing the sensor by 3 in each direction to represent "true pixels" whereas a normal sensor divides it by 2 in each direction. Any resolution above these divisions is accomplished 100% through interpolation and that has a bigger impact on the results at this stage than any other factor, given the same resolution sensor. That's why the rendering from the Pentax looks superior to the Fuji. In reality, it's a higher resolution camera by about 50%.
 
+Tim Bray First of all, the K-3 has a 24mp sensor (produced by Sony) giving it the exact same pixel count as the A7. The differences you see between the two primarily come down to the difficulty of optical design for smaller sensors. It's a great example of why full frame remains relevant. Optical designs become easier as sensors become smaller past the point of about 1" diagonal btw, which pretty  much puts Micro Four Thirds in the worst possible spot. And optical designs only become "easy" at the 35mm size, which is part of its original selection as a favorite compact format size.

As for switching to RAW, now you are simply comparing the different ways that Lightroom renders images for each camera. Yes, Lightroom handles each camera separately and differently with different uncontrollable parameters for base sharpness and other factors. Therefore it is a comparison of Lightroom's behaviors, not the camera. Especially in Fuji's case, Lightroom produces poorer detail results anyway. I aim to give the Fuji the best possible results when showing these comparisons.

Then referring to the green fuzz comparison, you may need some assistance to see what is lost. Take a look at this image: 

http://cl.ly/SfNj

What you may not notice until shown, but, should see glaringly afterwards is the X-trans sensor's inability to differentiate effectively between different intensities of green in all or mostly-green content. This is exemplified here by the interspersed light/bright green fibers mixed within the rest of the darker green fibers. On all examples except the Fuji, these fibers, though reorganized differently, stand out with good contrast and provide a texture that the Fuji fails to capture. As far as the practical implication this has, in the field, it means that foliage with a large gamut of green, foliage that gains its contrast due to green-yellow stripes and such, are not rendered as richly in the X-trans cameras. You likely won't notice it without a comparison, but, it's a shameful loss regardless, especially when the entire point of the technology is increased detail resolution don't you think?

Well, regardless, the facts are apparent. There are certain problems imposed by the X-trans design. Whether they affect you or your photography is entirely for you to decide. In every case, the problems with the X-trans are content dependent. My solution when using the Fuji X-E1 was simply to treat all of the images as though they were lower in resolution, reducing maximum print sizes, and though in some cases it wasn't enough (muddy trees, for example), it worked out okay. Usually the vibrant color rendering from the Velvia film emulation helped distract from any detail issues anyway, leaning more on color than detail. Furthermore, the problems were less obvious when there wasn't so much green in the image. I was able to produce some of my favorite images with the X-E1, just as long as I didn't want to view them at very large sizes.

Not everyone prints big. I do. I love a good 30x20. I love to stand far away and enjoy the composition, then get closer and see what's going on, then get closer and read the words on the sides of the 1cm semi trucks rendered in the image. But that's me, and that drives the attention to detail here. And obviously, if you're not on that same page with me, you may as well just skip any of the technical comparisons I make as they aren't serving you.

But, you're not doing anyone any favors by just dismissing objectively factual behaviors of the tools an artist might choose to use. It's always better to know the weaknesses so they can be compensated for appropriately.
 
+Alessandro Michelazzi The performance at 6400 is largely indistinguishable without software analysis, honestly. The details can be better coaxed out from the higher resolution cameras, but, for typical sized prints/outputs, it's not an especially relevant area of comparison any longer. In some cases, the results will differ in ways that only a subjective choice can decide what's best.

So I'd say that, depending on the exact content and particular desire of the user, one of these cameras may provide a better outcome than the others and it's not the same camera in every case. But I would also say that all of the camera provide an acceptable result, even into the 12,800 range, for appropriately sized prints or digital files.

In short, they're all just pretty great at high ISOs.
 
"In fact, that's the true reason to be enthusiastic about the Fuji X system, it has by far the best APS-C lenses made to date. The other reason is the incomparable image quality coming right out of the cameras." ~Zachery J

"Right now, if you want a modern lightweight camera (aka mirrorless) and care about optical and image quality, Fuji is winning. At least until the Sony A7 series comes out, but, even then it will take a year or longer for the lens offerings to catch up." ~Zachery J

So what's your point now?

I'd expect full frame cameras with 24mp and 36mp to have better IQ than a 16mp APS C. My D800 has better IQ than my Fuji's too. 
 
So wonderful to read your fine analysis. I had no idea my fuji xp1 was taking such terrible pictures. I'm going to list my entire fuji x kit on eBay immediately after posting this comment. I'd really appreciate hearing your recommendation for a camera system tha passes your rigorous tests.
 
Gene, no need to be nasty. Obviously, Zack is very particular and there is nothing wrong with that. I appreciate all the work he does in comparing and researching. 
 
My question to Zach is whether he feels that the differences between the cameras is noticeable at viewed resolution rather that zoomed way in. Zack, is this something that most people would notice at all or would it only be a person with a trained eye looking very closely?
 
+Gazzaroonii T. The image quality is quite high, if you recognize the resolution limitations. The green problem probably the most glaring issue but it is very situational.

The the reason to compare with the A7 now is the price. X-Pro1 launch price is the same as A7 launch price. Lens spread is similar, though a different selection of actual lenses. The lenses are a little more expensive on the full frame E mount line up, but, not that much. The point is that Fuji said they don't need to make a full frame camera. Now the rumors are that they are definitely making one, a fixed lens one next year and a system the following year. I think they recognize the problems with X-trans and Sony's move here just further legitimizes the market for full frame mirrorless. And the last point, really, is showing how an anti-aliasing filter on the same total pixel count can produce better quality images and isn't necessarily an evil thing as a recent trend has begun to imply.

But, more importantly, I have spoken in the past about certain image issues with the X-trans sensor and I wanted to give a very clear example of the limits in the technology's design. With so much going for the system, these warts are very frustrating to me.

I have a 20x10 print here made from an image straight out of the X-E1 (besides crop) that I really love. It's a beautiful evening city skyline, with wonderful colors. There is no doubt that great results can be produced with the camera. It just pays to know what it really does to get the best final results.
 
+Darren Humphries Most of the time, it won't be a noticeable problem. If you take that example link and switch to "print" size at the top right, you'll see that it's much harder, if not impossible, to see the difference. That normalizes all the images to 8mp, which is suitable for an 8x12 print. (Okay, I can still see it, but, in an actual print it will be very hard to notice.)

Where these problems manifest the most is for larger prints where you're stretching the 240 dpi recommended target or totally destroying it. I have a 30x20 from a 12mp camera that I like, but, I can see all the interpolation done to get a print that big from a file that low in resolution. Not everyone's going to notice. Not everyone's going to care, of course.

And at 12x8? Even 18x12 (with some extreme case exceptions)! You probably don't have to worry about the false detail, especially if you have nothing to compare it to.
 
+gene lowinger I'm fairly certain that nowhere in my post had I advocated this as a reason for anyone to go sell their camera. Nor had I ever said that these cameras produce terrible images. They have certain limits that are antithetical to the stated goal of their sensor design. This is a demonstration, hard evidence, of one of the two main limits that will govern the maximum capabilities of these cameras.

No serious user of tools in any other practice or profession would disregard knowledge about those tools in favor of sheer tribalism in the manner that photographers do. In fact, it almost seems as though you've taken personal offense to these results, which is strange as I'm fairly certain you're not the person who created any of these cameras. Interesting indeed.
 
Im pretty positive 3-pass demosaicing wouldnt cause this issue.
 
It's very strange to me that everyone seems to think that any camera rendering isn't enhanced in some way whether it be a Canon, Nikon, or Fuji, a digital image is always going to be only representative of what it is replicating et al digital versus analog music, it all depends on how the image or sound, or whatever is processed. Fuji's X-Trans Sensor is new technology as such Older testing and developing methods may not recognize the results as valid (it's digital therefore the software algorithm used must understand the data). But my eye see's what it see's and the images produced are stellar even if DXO doesn't say they are who cares. Just keep shooting and enjoy the photo's you produce, and one day magically DXO or DPReview, or whoever will find that Fuji's APS-C X-Trans sensor is really really good, even though it doesn't fit the mold, guess what neither do we...
 
It is an interesting comparison. However:

If you change the sample cameras to other 16.3MP cameras, the differences reduce significantly, if not tip in Fuji's favor. I'm looking at the Pentax K500, Sony NEX5T, Sony NEX 6, and Sony SLT A57 and I'm not seeing anything markedly better than the Fuji results, and to my eye, some of these (NEX 5T, Pentax K500) look a little worse. Still, all within the margin of acceptable relative to one another.

But then I move the target from those diagonal lines to the resolution targets made of converging horizontals or converging verticals. And that's where I find results that are a little less vague. Compared to its 16.3MP brethren, the Fuji camera absolutely destroys the competition here. For one, many of the cameras (the Sonys) display an unacceptable amount of false color moire. For two, the Fuji outresolves the others by a very perceivable margin, continuing to resolve fine detail well beyond where the others stop.

But one more test: move the target to the Siemens Star and notice that the Fuji reproduces this perfectly (the Pentax does a good job of it too). The Sony examples, for reasons I cannot discern, create an odd "pincushion" artifact near the center of the star. It's a weird thing that I certainly cannot account for.

All of this to make this point: your original conclusion is that the X-Trans sensor design is flawed, and that the poor performance on the diagonal resolution target proves that the X-Trans design is flawed. But eliminating other variables (roughly, both pixel pitch and sensor size... all of these sensors have 3264 photosites in the vertical, and within 32 photosites of one another in the horizontal and are within .1mm of one another in any dimension) shows that at worst (the diagonal lines test you chose) the X-Trans does nearly as well and at best (horizontals, verticals, siemens star) it handily bests its competition. Hardly conclusive proof that X-Trans, rather than pure resolution or photosite size, is to blame here.

And a final thought.... comparisons like these don't offer much in the way of real world applicability. I rarely encounter tight diagonals in the built world and never in the natural world; good for Fuji. I also don't find many Siemens stars in the real world; good for Sony.

But if I had to rely on one of these bizarre tests, regular verticals or horizontals are the thing that I do occasionally encounter in the city. And the Fuji clearly has the advantage there.
 
It's a nice tool. It's good to use it for comparison. When used effectively it helps to see what different cameras resolve in different situations.

Your conclusions are, to be honest, flaky at best. The only thing of note is that the IQ of FF 24mp and 36mp image sensors are better than a 16mp sensor.

Your sudden leap that this comparison proves your amateur theories, which so far seem to be yours alone, is without credibility.

You credit people opposing your point of view to some kind of fanboism or tribalism in above, maybe that's your poor judgment. I'm sure there's plenty of that fanboism about, but I'd say it's not the general consensus here. Your inflammatory language in the original post, inconclusive evidence and leap to poor judgement, is what's at the center of the responses I'd guess, certainly the root of my disagreement with your post.

I'm sure Fuji will do FF. The rumours are usually accurate. I doubt they'll make a 36mp one to compete with a 24mp bayer sensor, because, well, that's just silly talk.

While I wait for Fuji, I may get an A7 and scrap my D800.
 
I personally also shoot a lot of Leica M glass, and one thing I really like about the Fuji X series (read, X-Pro1 and X-E2) is the ability to use my Leica lenses on the bodies.  With both Fuji and Leica lenses I have gotten outstanding prints at 13x19.  So, I don't know where the A7 will stack up for me..
 
+Tim Bray Great comment, Tim Bray: "On the other hand, if slightly-angled fine black lines at 100% are what matter to you, those Sonys are hard to beat. "

The author's emphasis and motives of their post gives too little attention to (1) at low ISOs, all the cameras work pretty good, so who cares about the differences and thus (2) since anything works OK in bright light, the thing a demanding user would care about, is how well does the sensor work under difficult/low-light conditions? And I like looking at the Fuji ISO 6400 photos a lot more than any of the other low-light photos DPreview shows in its comparison.

Why are any of us wasting editorial space looking for negatives? For goodness' sake, if you HAVE TO LOOK for the negatives, it means those heroically-noticed negatives are not important. Who cares if you found that the Emperor has clothes, but upon careful scrutiny is not wearing underwear?
 
Zachery, take a look at fuji shutter speed. It's always slower than on competitors.
IRL it means you will end up with higher ISO on Fuji. You can also check DxO Optics 9 with its brilliant PRIME NR. Using it with 16mp cameras (nex, nikon, pentax) you will get at least as good noise performance at equal nominal ISO values with Fuji (even better in fact). But once Fuji's sensitivity values are actually lower, bayer filter cameras become better high iso performers: xtrans is not supported by this software.
BTW, xtrans is not about high resolution (bayer destroys it on base isos), nor the king of high isos (it is more software dependent). It's just about quite effective and easy to compute NR, that let to use it in camera. Nothing over that.
 
It's very trying to see people still judging x-trans based on Adobe software conversions. Compare with Capture One please

Yes the xtrans arrangement has its compromises - as does traditional Bayer- but you do get more detail than Bayer with AA with the right raw converter.

As for moire it is true you can still get it but also true its significantly reduced vs traditional AAless Bayer.

Its pretty funny when you think about it right? fujis research team will go with a solution and waste cash on something that has no benefit. Right ;-)
 
@ Dennis - complete nonsense that xtrans isn't about higher resolution given you remove the AA filter. It does show a resolution increase over a sensor with an AA overall.

At a minimum you should see such increase easily for bw photography.
 
+Ricardo Hernandez you are mistaken. This is a comparison of camera produced JPEG files. And frankly, the X-E1 produces much higher quality JPEG results than any standalone RAW converter, overall, including capture one.

 
+Ricardo Hernandez it's true that Fuji claims higher resolution due to a lack of AA filter. However, this is only going to be the case under 100% ideal demosaicing conditions. That is not usually the case so in general, the final detail isn't any better than a normal AA-covered sensor of the same resolution and at times, demonstrated by my post, the result can be even worse than low detail, it can be FAKE detail which is basically the same as moiré but without added color.

As for B&W, there is no detail gain due to shooting B&W. The whole sensor is covered in a color filter array and light intensity varies across the color spectrum, so the camera has no choice but to convert full to color before then converting to B&W.
 
Your post doesn't demonstrate anything

First it is false you can't get more detail with capture one over fujis jpeg engine. I agree te jpeg engine is fantastic but I routinely get more detail in a way that can e seen with capture one conversions

Two- you pick one area of the shot and there are other areas where te fuji does better. That hardly proves the point.

The lack of AA is not a fuji claim- it is a simple mathematical and empirical fact that an AA filter will introduce loss of detail. Without it the Fuji will have more detail I. What samples to the phosites. Certainly not all shots areas will show an advantage to xtrans but many will certainly do


 
It's clear and obvious you have no understanding of how a modern CMOS color imaging sensor utilizing a bayer (or Fuji's modified bayer) color filter array works. You can educate yourself on this concept and the impact it inevitably has on detail rendering on wikipedia. But suffice it to say, using 3x3 grids will definitely result in lower initial accuracy than typical 2x2 arrays "mathematically." It's unavoidable.

And if you can't see the painfully obvious rendering of non-existent line segments in the sample, you've got no sense of detail and this post isn't for you anyway.