X-Trans vs Foveon — A (Mostly) Monochrome Photowalk.
I’ve wanted to compare the monochrome monster — the Sigma DP2 Merrill — against the Fuji X-Pro2’s “Acros” film simulation mode. I’ve been in love with the Sigma monochrome output since I first picked up a Sigma camera back in 2014 or so.
Fuji’s much touted X-Trans sensor and film simulation modes have finally made their way into my photographic toolset, and while I love the output of my Fuji cameras, I don’t know that they’re as capable at monochrome images as my Sigma cameras.
My general sense was that the the Sigma Merrill cameras would have a smaller dynamic range (owing to their older & more niche sensor technology) and Fuji would have a wider dynamic rage since it’s a more modern sensor & benefits from the large amounts of development CMOS has had in general.
Also just from personal experience, I expected Sigma’s mid-tone tone curve to be steeper than Fuji’s.
Finally, I expected that Sigma’s microcontrast would be better than Fuji’s. Sigma’s, again owing to their unique sensor technology.
I discuss both sensor technologies in more depth here if you want to geek out a bit.
I used the Fuji 27mm f/2.8 lens, which gave me about the same field of view as the DP2 Merrill (which has a 30mm f/2.8 lens) — both on APS-C sized sensors. Normally I would have packed a tele lens for the Fuji just to give myself more options, but I wanted this to be something of a head-to-head comparison — being able to take roughly the same photo of the same thing with each camera.
That said, the 27mm f/2.8 isn’t Fuji’s sharpest lens and the lens on the Merrill is very sharp — but it would be unfair if all of the Fuji images were “zoomed in” (telephoto) and the Merrill’s were not.
(On second though the Fuji 35mm f/2.0 would have been a better choice for sharpness. Oh well.)
I also brought along my Ricoh GRD1 — legendary for its grain structure at high ISO — “it looks just like Tri-X” according to some. Most cameras have horrible “noise reduction” algorithms that make photos look smeared and gross at high ISO. The 2005 Ricoh GRD1 has a max ISO of 1600, and many aficionados push it to ISO 1600, shoot it JPG only mode, and leave it there. So that’s what I did.
The weather in New York was a bit schizophrenic — rainy one moment, sunny the next. Sometimes rainy and sunny at the same time. It made me a little paranoid about getting my cameras wet — I trusted the Fuji in the rain more than any of my other cameras… I’m not sure if it was the talk of the “Weather Resistance” on the Fuji or the fast autofocus — but when I had my umbrella up (the ultimate one-handed operation test) — it was either the Fuji or Ricoh that was in my hand. One with fast autofocus — one with snap focus.
The Sigma would have been slower to autofocus & something about it made me worry that a drop of water in the wrong button would kill it — I’ve had it happen to me before with some of my {brand unspecified} cameras. I used to shoot a lot at the beach & one of splash of salty + sandy water has killed more than one of my cameras. I think one summer I lost 3 or 4 cameras to the beach.
That said, the relative lack of grip & the weight of the X-Pro2 made it uncomfortable to hold for long periods of time while walking around. My Sony A7 has a better grip and is a joy to hold — though the later models suffer from a slightly larger body leading to an ergonomically worse grip — for some reason when Sony added In Body Image Stabilization, they made the camera a bit larger in all the wrong dimensions leading to a more uncomfortable grip.
One of my favorite things to do in Chinatown is to just get lost. I’m a New York City native — I know most of NY like the back of my hand, but when you get downtown to the Chinatown / Financial District / Lower East Side area, streets start to intersect at weird angles and it’s easy to get lost, which is exactly what I did.
Chinatown and the Lower East Side and NYC’s Financial District all sort of meld together at some point, and that’s the beauty of New York City. People, cultures and cuisines commingle seamlessly. In some ways, it brings me back to the 90s — before we had restaurant reviews in our pockets and we’d decide what to eat based purely on what the restaurant looked like & how many people were inside. Back when I’d show up with a camera anywhere and it was both special because I had a camera, and also nobody cared because there was no “online” for their photos to go, so everyone was free to be themselves.
I have an upcoming project that I want to shoot in Monochrome. Something that’s really interesting to me personally, so I want to find the tool to take the photos. I know the Sigma produces amazing monochrome images… But it’s very slow. So how does the Fuji stack up?
Which camera would I trust to save these images for posterity?
It’s worth noting that I owned a Sigma DP3 Quattro and a Sony A7R2 — both of which are excellent cameras, but I ended up selling them. Technically, on paper, they’re better than the Sigma DP Merrill and Fuji X-Pro cameras I have now.
For me, the Merrill sensor has more personality than the Quattro, and the Fuji looks better with less work than the Sony. It was a difficult decision, but I went for lower megapixel count files that I preferred working with.
I’d rather turn over an SOOC file that I’m confident with little to no post-processing than slave for hours trying to find colors I like on a file I don’t like.
A Technical Note & RAW Files
Medium (the website that hosts this blog) resizes landscape images to 2000px or so wide and then displays them at an arbitrary width forcing your browser to resize the images. As you go through the rest of this blog post, I’ll be showing images side-by-side.
As much as possible I exported to 2000px to limit how much Medium’s JPG compression algorithms messed with the images, but I’m letting you download these images so you can compare them yourself — both JPG and RAW files. (If you want TIFF files it’s up to you to create them.)
My stated goals may be different from yours, so I’m sharing these files with the community via a Creative Commons Non-Commercial, Attribution license. Meaning you can use the files however you see fit — as long as you don’t make money off of them (so don’t post them to your blogs if you have affiliate links — that means your blog is a business) and as long as you credit me by linking back to this article.
I’d also love it if you came back to this article & commented on it with a link, so anyone viewing this article has the opportunity to explore whatever else comes from it. Not that I think I’m doing anything special other than having two cameras. I would make this part of the license, but Creative Comments doesn’t have a “Non-Commercial, Attributional, Comment Back” license, so you’re on the honor system for that part.
I’ve made available all of these RAW + JPG files, as well as 7 other comparison images for your pixel peeping pleasure.
The comparison files are available via this Dropbox link.
This is the DP Review discussion I started about this comparison, go there for a more learned discussion of the technical aspects of the two cameras.
So — consider this your “official” warning — my opinions are my own & may not be yours. Small web image aren’t the best for pixel peeping quality, and I do some funky resizing of images during the comparison. If you dislike these methods — you can (and should) download the files yourself.
(also if you can, set your browser to 2000px wide since that’s the width of the files being viewed.)
A Random Chapel
Whenever a friend from Europe comes to New York and I have to play tour guide a bit, I end up showing them buildings and describing their history and then realizing that a 200 year old chapel is nothing compared to what they’re used to.
I suppose that’s as good a metaphor as any for this comparison — two relatively upstart or at least niche camera technologies, duking it out in a world of Bayer sensor giants.
I had two criteria for this comparison.
First — At standard viewing sizes, is there a camera whose rendering I prefer, since that’s how most people will view the photo. Second — On a pixel-peeping level, is there a camera whose rendering I prefer, because that’s how I will have to edit the files.
For purposes of both expedience and posterity — is there a camera that I felt was better suited to the job?
Both are subjective measures. I’m not doing any sort of technical analysis — I’m not that sort of photographer. I take photos because I enjoy it & I edit them mostly based on what pleases me. Again —this is not a scientific comparison, whenever I use words like “Accurate” — it’s not a technical term, it describes a feeling — much like a story can “accurately” describe an emotion, these impressions are meant to be personal only.
I’ve been staring at these photos for about a week now — making minor edits to the blog post for some reason — obsessing over it before I hit publish*, so I’ve come back to this “first impressions” photo a few times, and on this level, what’s striking to me is how Foveon-like the Fuji sensor is.
* probably in anticipation of the backlash if I get anything wrong… — also it’s taken me 6 months to finally finish up this post.
I usually associate the Foveon sensor with a “3D Pop” that’s lacking in Bayer sensors (and I believe lacking in the 16 megapixel X-Trans sensor of the X-Pro1, and to a lesser degree in the Foveon Quattro sensor). A sense that each pixel is located in 3D space, not just on a 2 dimensional plane.
Though the X-Trans image seems overly sharpened — algorithmically enhanced — the Fuji image has more 3D pop than I see from Bayer sensor images (unless a lot of sharpening is added in post).
Upon first seeing a Foveon Merrill image, many people’s initial reaction is “is that an effect? Because that’s not what pictures usually look like.” And having used Foveon sensors for a few years now I believe — yes, it is partly an effect — choices made by Foveon and Sigma engineers to make their photos look a certain way by applying algorithms to the images. But also it’s partly the removal of the “Bayer blur.”
The Fuji feels as “3D” as the Sigma at this level, and that was surprising to me.
Straight Out of Camera Comparison
Both images taken Straight out of Camera, composited in Photoshop & cropped to 2000px wide.
It’s fascinating how different the Fuji rendering is from the Sigma rendering. To make a crude analogy, the Sigma feels like a charcoal drawing of the scene with lots of grit and detail. The Fuji feels like an oil painting of the scene with an emphasis on shapes over textures.
I suspect a lot of this comes from how the sharpening algorithms work on each camera. The Sigma seemingly prefers “microcontrast” — looking for where adjacent pixels are different from each other. The Fuji seems to prefer edge-detection — looking for where adjacent pixels are similar or different in an attempt to find shapes. (See discussion of X-Trans “wormies”). This is an oversimplification of these concepts.
Enhance 15 to 23. Give me a hard copy right there.
The 16 megapixel Sigma file is considerably smaller than the 24 megapixel Fuji file, in terms of actual pixels in the resulting JPG.
Sigma claims that its 16 megapixel Merrill sensor is really a “45 megapixel” sensor and “equivalent to” a 30 megapixel Bayer sensor, having to do with the complex way the pixels are laid out. (geek out on this here.) Long story short — the Sigma Merrill file should survive upscaling better than a Bayer file, and Sigma’s raw editing software — Sigma Pro Photo (SPP for short) — even has an “export to double size” option that turns the 16 megapixel file into a 64 megapixel file (it doubles the output in both dimensions).
Below is a file from the Sigma that I up-sized using Sigma Pro Photo’s “Double Size” export and then downscaled to 24 megapixels.
Sigma Workflow: X3F (RAW) file > SPP (defaults) > Double Size Export, 16 Bit TIFF (64 megapixel file) > Photoshop > Resize to 24 Megapixels (“Automatic”)
I know some of you will argue with me over this method— it’s not fair to the Merrill to have to be uprezzed in Sigma Pro Photo just to be downrezzed in Photoshop. To which I say — yes, that’s totally a fair criticism & to which I say, you have access to the RAW files. There is no such thing as a RAW to JPG conversion that doesn’t have at least some opinion on how it should be done, so if you don’t like mine — do your own. What I’m aiming for is transparency and repeatability — I tell you my steps, and therefore my biases.
Not much changes in this version — the Fuji retains the grit I spoke of before, and maybe shows a touch more detail. At this point — I’d say I prefer the Sigma Merrill image to the Fuji, even if it feels a bit less sharp — but I suspect that feeling of sharpness is about sharpening algorithms & not what the sensor is seeing. The Fuji feels sharp — but it’s sort of a cartoonish version of sharpness — again like an oil painting of the scene.
Detail:
Both Compared from RAW
OK — so I’ve been complaining a lot about Fuji’s RAW conversion algorithms, specifically when it comes to sharpening. So let’s convert the Fuji file from RAW.
Raw Therapee is said to have the best Fuji X-Trans conversion algorithms. The output without any sharpening was soft, so I enabled a few of the sharpening options — Surely the Sigma image has some sharpening applied (In Monochrome mode the slider is disabled, but it’s more than likely that their software applies some sharpening algorithms).
The limits of the lens are apparent here — the areas towards the top of the image (which were closer to the edge) appear softer than the areas towards the bottom (which were closer to the center). It’s interesting that this wasn’t really evident until now & points to heavy-handed sharpening and lens correction on the part of Fuji . Though looking back I can see how the SOOC images are a bit soft there, but the impressionistic “water-color” like rendering of the images hides this.
The “oil painting” sense is gone here and it looks much more like the Sigma. Indeed, impressively so.
Overall, thoughts.
So — let’s go back to my initial purpose for doing this test.
I have an upcoming shoot whose output is monochrome. I’ve decided that it’s worth investing time and a modest amount of money to get the best image quality. On pure image quality, the 16 megapixel Sigma was — to me — the best option I had. The 24 megapixel Fuji, second best. The Sigma is slow — slow to focus, slow to write files. The Fuji is fast, but am I losing much quality by shooting the Fuji instead of the Sigma?
That is — if there’s a living, breathing person in front of me — is the advantage to the slow but more or less monochrome Foveon sensor worth the extra file write times & how that can ruin spontaneity?
Conclusions
I had the general sense going into this test that I would walk away with two very different cameras.
I tend to think of cameras and lenses as having personalities, and I reach for a specific camera + lens combination at any given moment because the scene I want to capture matches the personality of that combination.
I know that I have some of my own personal bias in this choice…. that I don’t always test all cameras in all situations, so I tend to gravitate towards certain cameras over & over again — like I prefer the X-Pro1 for natural light & the X-Pro2 for studio lighting. Why? Just because that’s how I first approached each camera & my impressions of how each renders a scene comes from that initial biased impression.
Still, I walked into this experiment with a very clear idea of what would come out of it. So let’s revisit my initial assumptions.
- The Fuji would have a larger dynamic range than the Merrill.
The Fuji probably wins here, but I wasn’t really challenging them so in terms of these photos, they feel pretty similar. - The Merrill would have a stronger tone curve & the Fuji would have more subtle midtones.
Eh.. Maybe? Going in to this, I thought I could detect the Merrill by tone curve alone, with Sigma favoring a strong “S” tone curve and Fuji preferring a more subtle mid-tone roll off. I would say that’s true, but not as pronounced as I’d assumed. In retrospect, maybe this is partly down to me thinking that some of Sigma’s “microcontrast” is down to tone curve. Sigmas also tend to under-expose in auto mode to protect the highlights, leading to more “crushed” blacks, again leading to my impression of a stronger tone curve on the Sigmas. - The Sigma “microcontrast” would be far superior to the Fuji — but the Fuji would be better than a Bayer because of the X-Trans sensor layout.
I would say this is true. (nerdy discussion here) The default Fuji “Acros” rendering does seem to be doing something weird with edge detection, which mostly goes away when edited through Raw Therapee (comparing the Raw Therapee unsharpened + sharpened + Fuji SOOC files is instructive here). While I didn’t compare a Bayer sensor camera in this test, my experience with Bayer says that it would be lacking compared to either sensor in this regard.
One of the reasons I wanted to do this test — I’ve always felt the 24 megapixel Fuji X-Trans sensor was the closest to a Foveon I’ve ever used — again due to the unique layout of green pixels & whatever Fuji algorithm sauce goes into demosaicing.
As for which camera I’ll decide to use for this project — I’m still undecided. I guess I’ll let you know when I actually complete the project.
[edit: It looks like I’m leaning towards the Sigma Quattro or Fuji X-Pro2, depending on how fast things need to happen.]
As for which camera setup is going on my next photo walk? usually it’s the Sony A7 because half of my purpose for doing photowalks is to get to learn the character of whatever lenses I own.
A Totally Non-Scientific Theory About Why They’re Different (and how they’re similar)
Having looked at the SOOC X-Trans file, the Raw Therapee (without sharpening) and the Raw Therapee (with sharpening) processed raw files — all available via the Dropbox download (along with a handful of totally different comparison images) — these are my observations on the difference between Foveon and X-Trans (and Bayer). I’ve also looked at Sigma files processed through Kalpanika — so without Sigma’s sharpening algorithms.
I’m not an engineer, these are just the impressions I’m getting from looking at these images — these are not meant to be technical discussions of why the two cameras are different, but I’m going to use technical sounding terms to describe my impressions.
The Sigma Merrill sensor + processing algorithms tend to prefer “microcontrast” — seeing what’s different between adjacent pixels and increasing the contrast between them.
The X-Trans sensor + processing algorithms tend to prefer “edge detection” — looking for shapes and then smoothing out the details within the shapes.
(I suspect this is part of why Fuji images have “wormies” when standard sharpening algorithms are applied to them.)
So while the Sigma Merrill resulting JPGs look like — almost like each pixel is its own grain of sand, the Fuji X-Trans is purposefully looking for areas of like or dislike, leading to an “oil painting” feeling — areas of contiguous shapes with less detail within the shapes themselves.
This “oil painting” — sometimes smooth and sometimes sharp — feeling is Fuji’s version of Bayer Blur. Where a Bayer CFA is all blur — no adjacent green pixels — the Fuji is a combination of both adjacent green pixels & not adjacent green pixels. To produce a pleasing image that’s not overly micro-contrasty, they apply some sharpening and some smoothing at the same time.
Running the X-Trans sensor through Raw Therapee, rather than Fuji’s default JPG algorithm (at default sharpening levels), the X-Trans begins to look much more like the Sigma Merrill — the “oil painting” feeling is gone and the pixels seem to have more of that charcoal drawing “grains of sand” quality to them. It’s still “smoother” than the Merrill but it’s much more Foveon-like than the default JPG rendering.
I remain very impressed with Fuji’s X-Trans sensor, even if I am a bit disappointed in their JPG rendering engine and the way it creates an impressionistic “oil painting” version of the scene.
