X-Trans

[wblink]

I am using (very happy with it!) a Fujifilm T-X1.

I am NOT happy with the way LightRoom
handles these (X-trans) files.

It is different from BAYAR sensors, but they are or exiist in any way, just different.

Please Adobe: attend to this ......

 

[Jim Wilde]

Willem, there's nothing we can do about the issue here....we are not Adobe.

Suggest you add your voice to [url="http://feedback.photoshop.com/photoshop_family/topics/fuji_x_trans_support']this thread[/url] at the official Adobe web-site.

 

[Modesto Vega]

There is also an active thread of the Adobe forum - https://forums.adobe.com/thread/1493324?sr=inbox&ru=2213492.

IMHO, this is not just limited to Fuji, I have encountered similar problems with my Nikon's. The solution I have identified with some very valuable help from regular contributors to this forum is to make use of Camera Calibration profiles. I envisage this as having a number of steps. The 1st step, where I am right now, is to try to understand better what the different profiles supplied by Adobe do. This is leading to ditching the "Adobe Standard" camera calibration profile as the basis of my workflow.

The next steps involves creating profiles for my Nikon D600 under certain settings.

 

[johnbeardy]

You know, I think the Fuji issue is more fundamental as it seems to relate to demosaicing the raw data from the very different xtrans sensor. Aren't you really talking about colour rendering, which can usually be resolved by Camera Calibration?

 

[Modesto Vega]

You know, I think the Fuji issue is more fundamental as it seems to relate to demosaicing the raw data from the very different xtrans sensor. Aren't you really talking about colour rendering, which can usually be resolved by Camera Calibration?

Am I talking about colour rendering? Possibly.

But could I please ask you to consider things from a different angle? Just to get something clear on my head and, hopefully, to help others.

X-Trans is a type of CMOS sensor with a Bayer filter as the OP mentioned. It uses an irregular pattern of pixels to reduce moire without needing an anti-aliasing filter. Apparently this arrangement is similar to Silver Halide film. It basically uses a 6x6 repeating pattern that goes like

GBG|GRG
RGR|BGB
GBG|GRG
-----------
GRG|GBG
BGB|RGR
GRG|GBG

So I am expecting a RAW file to contain similar data to the RAW file I get for my Nikon D600 or the RAW file coming from a Canon EOS 5D Mark III. In a nutshell:

• A short file header containing the byte-ordering, a file identifier and offset into the sensor image data
• Metadata from the camera sensor, including the size of the sensor, the attributes of the colour filter array and its colour profile
• The sensor image data (which is the RAW data) - this is going to contain readings for the red, green, blue and (second) green filters of the colour filter array

AFIK Lr discards most of the camera sensor metadata and this is a problem with certain settings available with the X-Trans sensor, such the various DR modes. I would not be too surprise if it does something similar with the Nikon Expeed 3 sensor.

If you look at the examples attached to those 2 threads, you are looking at completely different interpretations of the RAW data. It cannot be just demosaicing, there are significant colour changes. The colours look too different, just look at the greens.

The point I am trying to make is that Adobe appears to have made an attempt to shoehorn all sensors and all raw files into one single standard, the "Adobe Standard" which is the DNG profile Lr uses by default to tone map and make decisions about which colours to render. I don't have a problem with that as long as the side effect is clearly stated but the side effect is not clearly stated.

What is the side effect? Lr might not produce a "correct"/"true" reproduction of what the camera took (please note the quotes) and that in order to get a "correct"/"true" reproduction a different or custom DNG profile might be needed. Once that side effect is clear, Lr is a smashing product.

Just imagine for a minute photography reduced to Ektachrome film and/or Ektachrome film processing. The closest thing we have nowadays to different types of film are different types of sensors.

 

[sizzlingbadger]

Your correct, LR does try to make all raw files look the same, an "Adobe Standard" look. They also give you camera profiles that try to mimic the in-camera jpg files for some manufacturers like Nikon, Canon, Fuji, Sony etc. But you can create any look you want and save it as a preset so your creative options for "looks" are almost boundless.

Camera profiles do not help with the common x-trans issues in LR, it is the fundamental way they are demosaiced that causes issues. Sharpening settings can help to hide it but you cannot fix it. Many other raw convertors struggle with the x-trans raw data too. I have tried many convertors and wasted many hours but was never happy. It is more difficult mathematically to calculate the pixel values so they will always be slower (require more cpu) to render too.

If Fuji used a bayer sensor in their mirrorless cameras I'm sure they would sell twice as many of them, they are great cameras but the x-trans idea although clever on paper is yet to deliver in the real world in my experience.

After 2 years I finally sold all my Fuji gear (3 bodies, 7 lenses) and bought a Sony A7II, it solved all the issues I had and I'm very happy now. I have a 24Mp camera with smaller raw files, faster conversion and no artefacts that ruined my B&W conversions. The only downside is its a little bigger and heavier. The results I get in Lightroom are great and I like the Sony camera profiles but I have to give it to Capture One for the best conversion overall.

 

[Modesto Vega]

Your correct, LR does try to make all raw files look the same, an "Adobe Standard" look. They also give you camera profiles that try to mimic the in-camera jpg files for some manufacturers like Nikon, Canon, Fuji, Sony etc. But you can create any look you want and save it as a preset so your creative options for "looks" are almost boundless.

Camera profiles do not help with the common x-trans issues in LR, it is the fundamental way they are demosaiced that causes issues. Sharpening settings can help to hide it but you cannot fix it. Many other raw convertors struggle with the x-trans raw data too. I have tried many convertors and wasted many hours but was never happy. It is more difficult mathematically to calculate the pixel values so they will always be slower (require more cpu) to render too.

What I would love to know is if camera profiles play a role in de-mosaicing a RAW file or de-mosaicing takes place before camera profiles are applied. In other words, I would love to know if all the colour decisions Lr makes with RAW files is a 2 step, de-mosaic and calibrate - or 1 step process - just de-moisaic.

This is in a nutshell what I was asking John Beardsworth and others to consider.

 

[clee01l]

What I would love to know is if camera profiles play a role in de-mosaicing a RAW file or de-mosaicing takes place before camera profiles are applied. In other words, I would love to know if all the colour decisions Lr makes with RAW files is a 2 step, de-mosaic and calibrate - or 1 step process - just de-moisaic.

This is in a nutshell what I was asking John Beardsworth and others to consider.

De-mosaic is a separate process that generates rgb pixels from the photo sites. As has been described earlier the X-Trans photo sites are very different from the Bayer array. The result from either array is a red value, green value and a blue value in each pixel triplet. Up to this point no post precessing has been applied. Post processing is applied before the RGB image is released to the LR Catalog. There is some WB, Some sharpening and some NR along with Adobe's best recommendations for an initial image. Camera Profiles are applied (I think) in LR. These are for the most part mimics of the in camera processing. Nikon calls these Process Controls. Canon uses the term Picture Controls and I'm not sure how they are named by Fuji. Post processing modifies the RGB values. If you want to see what comes storage out of the demoniac process take a look at the General Develop preset called "Zeroed" It still applies a LR "process version" and a camera profile. But this is as close as we can get to SOOC for RAW images in LR.

it would be nice to be able to stop the conversion in ACR after the demosaic process and before and adjustments have been made to the derived RGB values.

 

[Modesto Vega]

Cletus - this is an excellent summary. Just a few comments.

As has been described earlier the X-Trans photo sites are very different from the Bayer array.

The X-Trans sensor is a Bayer filter, a modified Bayer filter, but still a Bayer filter, like the Foveon X3 sensor. As you confirmed what causes the difficulty is the very different arrangement of the photo sites. Intellectual property issues - i.e., patents that have not been granted yet - are the likely cause of why an algorithm has not been produced yet by Adobe for this sensor.

Post processing is applied before the RGB image is released to the LR Catalog.

This is the key part, how much post-processing on the RGB image does Lr or ACR do before the image is released to the LR catalog, before the image is released to the photographer for further manipulation? This is the unknown quantity with the X-Trans and other sensors. You are right the "Zeroed" preset is the closest thing to the de-mosaic file but the "process version" and a camera profile, presumably "Adobe Standard", have been applied which skews the results.

It would be nice to be able to stop the conversion in ACR after the demosaic process and before and adjustments have been made to the derived RGB values.

Yes it would be very nice, not because the images would look nice, but because it will help to assess what Lr is really doing to the RAW file when different settings are applied. IMHO, the true negative in the digital world is the de-mosaiced raw file without any post-processing whatsoever applied to it; it doesn't look pretty but film negatives did not look pretty neither; the true negative is not a RAW file post-processed with a set of parameters and algorithms Adobe deems to be standard.

 

[Modesto Vega]

Incidentally, if you look at the original front page of the patent application for the Bayer filter, you would notice an schematic of a 6x6 filter - http://pdfpiw.uspto.gov/.piw?Docid=...patft.uspto.gov%2Fnetahtml%2FPTO%2Fpatimg.htm. Very interesting!!

 

[Denis de Gannes]

Cletus - this is an excellent summary. Just a few comments.


The X-Trans sensor is a Bayer filter, a modified Bayer filter, but still a Bayer filter, like the Foveon X3 sensor. As you confirmed what causes the difficulty is the very different arrangement of the photo sites. Intellectual property issues - i.e., patents that have not been granted yet - are the likely cause of why an algorithm has not been produced yet by Adobe for this sensor.


This is the key part, how much post-processing on the RGB image does Lr or ACR do before the image is released to the LR catalog, before the image is released to the photographer for further manipulation? This is the unknown quantity with the X-Trans and other sensors. You are right the "Zeroed" preset is the closest thing to the de-mosaic file but the "process version" and a camera profile, presumably "Adobe Standard", have been applied which skews the results.


Yes it would be very nice, not because the images would look nice, but because it will help to assess what Lr is really doing to the RAW file when different settings are applied. IMHO, the true negative in the digital world is the de-mosaiced raw file without any post-processing whatsoever applied to it; it doesn't look pretty but film negatives did not look pretty neither; the true negative is not a RAW file post-processed with a set of parameters and algorithms Adobe deems to be standard.

Every raw processing software package uses their own proprietary processing of the raw data including the Camera Manufacturers own in camera firmware. There is no standard/ correct recipe, just different interpretations.

 

[Modesto Vega]

Every raw processing software package uses their own proprietary processing of the raw data including the Camera Manufacturers own in camera firmware. There is no standard/ correct recipe, just different interpretations.

I don't think anybody in the thread is arguing about correct recipes. Of course there are all interpretations but there are interpretations which are closer to creative intent of the photographer than others. Do you agree with this? Have I just gone mad?

As a digital photographer, few things annoy me more than having a creative intent, taking a photo or set of photos, doing a few cursory checks in camera to broadly confirm the creative intent, import the photos into Lr to find out that creative intent is not been rendered, and discover this is largely due to a camera calibration profile, I have been there many times.

Part of the creative intent of a photographer is the choice of gear, choosing an X-Trans sensor is a creative choice. Ansel Adams, Cartier Bresson or Dorothea Lange chose the equipment they used based on their creative intent. Just imagine Cartier Bresson working with a large-format camera, he would have gotten none of the wonderful photos he got. With this in mind, I don't see why support for the X-Trans sensor could not be improved, I am sure Adobe will address this in Lr 6, and why the whole interpretation process cannot be made more transparent, the process is not wrong, it is just obscure.

I hope nobody is arguing that because Adobe is such a big (and good) player, we all have to use cameras from one vendor or from vendors that work better with Adobe.

The process to develop Fujifilm Velvia film was clearly documented, E6, and the process was stable. I could do it myself or send it to a lab and would get consistent results providing the process was adhered to. It strikes me the equivalent to this in digital photography, which could be processing an X-Trans file, does not yet exist.

 

[Denis de Gannes]

Quote "I don't think anybody in the thread is arguing about correct recipes. Of course there are all interpretations but there are interpretations which are closer to creative intent of the photographer than others. Do you agree with this? Have I just gone mad?"

If your "creative intent" is what you see on the camera's display then the answer to your first question is yes. Some software packages will reproduce close matches to your camera display and the package provided to you by the camera's manufacturer will provide the best match. However based on what I have seen posted on many forums there are lots of users of Fujifilm cameras with the X-Trans that are not satisfied with the Adobe rendition of the X-Trans sensor. If I were using your approach then I would choose the raw converter that does the best job.

That said I am sure Adobe is trying to improve their rendition and hopefully this will come in an upgrade in the not to distant future. I say upgrade since I believe the problem is more than just creating / adjusting the profile.

I am an Olympus user and adopt a very different approach, I shoot raw and try to capture a properly exposed image and an appropriate composition for the intended image. Then I post process with the raw software of my choice to produce my "creative intent" (to use your phrase). I presently have Lightroom, Capture One Pro, Qimage and Olympus Viewer 3, which are all very good raw processors. Trying to match what I saw on the cameras screen hours or days before are the furthest thing from my mind.
In any event once you have imported the raw file into Lightroom you no longer have access to the camera thumbnail to compare while working with the raw file.

The answer to the second question is no.

 

[Modesto Vega]

If your "creative intent" is what you see on the camera's display then the answer to your first question is yes.

Oh well, I think I just gave a good excuse to call me mad, I hope you enjoyed it.

However, I think you missed the point. Creative intent starts with a choice of equipment - Olympus in your case, Nikon and a Sekonic Litemaster Pro in mine, FujiFilm X-T1 in the case of the OP. It continues with how the camera is set up, shooting RAW or JPEG - you the OP and I shoot RAW - and DR modes in the the FujiFilm X-T1 which Lr ignores. It continues with what type of lens is put on the camera body, how a photograph is composed, and moves into exposure settings. If I shoot with my Nikon D600, a Nikon AF Nikkor 20mm 1:2.8 D, set at f/2.8 and 1/4000 sec, that was my chosen intended "creative intent", forget the camera display if you find its mention distracting and excuse to call me mad , I would expect the software I use to interpret my RAW files to give me a reasonably good rendering of my chosen creative intent to start working with the photograph.

Lr cannot do that with a RAW file from an X-Trans sensor, we know that and Adobe knows that. With Nikon, to get the intended "creative intent" I very often have to change the Lr camera profile to get a decent start; so the "Adobe Standard" look does not work by default; once that is out of the way Lr does a great job.

Incidentally if a Fujifilm X-T1 is configured to shoot JPEG it comes up with photographs that look outstanding in Lr, the in-built software replicates the creative intent very well.

Lastly, the whole point of the X-Trans sensor is that it packs 6x6 photo sites per pixel, instead of the usual 2x2. In theory, it should be better at capturing colour, but if it cannot be interpreted it is a good as not having that capability. I hope for the shake of digital photography that handling different sensors and Bayer Filter configurations does not turn into a VHS or Beta situation.

 

[Denis de Gannes]

No problem, I agree with all that you have said, if Adobe LR / ACR does not produce an acceptable rendition for you then use the software provided by your camera manufacturer to render the raw data. This will match what you have expected, then you can make further adjustments to your likening.
From another point of view, why would you pay additional funds to Adobe to create the same rendition as the software provided by your camera manufacturer for free. If there is no benefit, real or perceived then why purchase the product.

 

[Denis de Gannes]

Just some further thoughts / points.
a. Processing of raw data produced by digital cameras is not a precise science, each and every processing software package uses their own proprietary processes which are not shared with other competing parties.

b. The firmware and software utilized by your camera manufacturer is unique and proprietary. It is not necessarily correct or accurate, color or tone wise and is designed by their engineers to present what they, perceive to be a pleasing rendition.

c. If you wish to test this concept, you can take a controlled picture of several inanimate objects with different tones and textures along with a Greytag Macbeth color checker. Render the raw file with different raw processors and view the results with the actual scene next to the computer. See if any match reality or match any of the other renditions. The results will surprise you.

example.
https://www.flickr.com/photos/baxter43/5355713456/in/album-72157625678052039/