Welcome the 14 BIT self-cleaning 2x RAW 1D Mark III!

[Paul Bestwick]

1d Mkiv

www.studio58.com.au

 

[Paul Bestwick]

1d 4

yes what you are saying is exactly along the lines of where I was coming from when I said I did not think that people fully understand the significance of this camera (newly engineered from the ground up)
Crazy as it sounds, like you I can't help but wonder about what the MKIV will be like. As for the DS series replacement......who knows what is coming there.
Having formerly worked for Canon during the period of the first digital cameras (kodak based) it is amazing to have been part of an amazing revolution & to be using the equipment during what really is the developement phase

Paul

www.studio58.com.au

 

[Ferenc Harmat]

1D MKIII implementation is a revelation, indeed...

...I did not think that people fully understand the significance of this camera (newly engineered from the ground up)

Absolutely. Not even a basica idea, in general, as well as the huge impact that this design and implementation will have on future Canon's models.

I am almost ready to predict, to literally state that this camera represents Canon's departure from what we currently know today, as Digital Imaging. It is closing a chapter in Canon's history, and opening up an entirely new one.

The devices that will follow hereafter (at least 1D-class), will respond to a dramatically different and ambitious vision of how we will capture, process and store media content, in general. All the seeds and elements necessary for the development and growth of these new breed of products have been carefully, almost meticulously planted on the 1D MKIII's design, and this is why this camera has such HUGE significance.

The MKIII also shatters that old concept about "PJ", "sports"-only equipment. You now have a camera that is capable of pretty much anything, and almost under any circumstance. And the way it does is with simple, pure engineering excellence:

• Instead of breaking apart its wonderful body into disfunctional or unreliable components, they went ahead and chopped/trimmed the frame and replaced the old battery, thus saving close to 225grams.
• Instead of falling pray to the "full-frame" cost-trap, they listened to their most important customers (those that buy these cams in LARGE chunks) and kept-and-improved the 1.25x APS-H sensor, now making it better, even with dust-reduction tools, on-board (ideal for those customers, indeed). It is substantially cheaper to produce than full-frame sensors, yet image quality is world-class.
• Instead of desperately increasing pixel count at the expense of sensel's sensitivity, the manage to pack an additional 2 Megapixels, thus providing the necessary resolution for just about most of the tasks, while still increasing performance accross the board, and also storage-efficiency with still-manageable RAW files. This, unfortunately, will not be possible with the next incarnation of 1Ds MarkIII series, and the 1D MarkIII draws a line that separates and differentiates 1D-class models, based on a choice of either uncompromised high-ISO performance, or uncompromised low-ISO resolution (this makes sense, though, and it is in perfect tune with the needs of each of these market segments).

From the above you can logically deduce that the next step for Canon would be to begin imaging FF sensors based on 1D MKIII technology. And that takes you back to 16 MP, which is where you are with today's 1Ds MKII. This means, to me, that Canon's most rational decision here would be to discontinue the 1Ds, as we know it, and strengthen the 5D-class with a better, grip-less, weather-resistant body, with better AF (no need for live-preview, high-speed dual-core pipeline, dual storage-ports, etc.), and same high-ISO performance as 1D MKIII, with 16-17 Megapixels, and simply stop there, in terms of resolution.

From here on, the company may move on with sensor development technologies, and spend more cash in the next league of sensors and optics that will enable an even wider range of performance and future products. If Canon wants to get into the MF business, and domiante it, it may better leverage on all of its sensor-making experience and accumulated R&D to produce foveon-like sensors that, with clearly lower pixels counts, will easily and comfortable compete with anything that is out there, with much smaller files, and also smaller optics, for instance... This could trickle-down later to consumer, smaller-format models.

Just some food for thought...


Ferenc

 

[Paul Bestwick]

Hey Ferenc,

so what are your thoughts re image quality.....specifically as it relates in comparison to the 1DSMKII.
I am thinking that the 10 MP beast may more than match it

Paul

 

[Ferenc Harmat]

Paul:

The 1D-class lineup is at a cross-roads, now, where further differentiation will necessarily imply a trade-off between high-ISO performance and Spatial Resolution.

The 1DMKIII is clearly going the high-ISO path, with uncompromised performance on this front, as compared to its 1D MKII-N sister. However, the 1DsMKIII will not be on-par inthis department, if it comes out with 1D MKIII's double pixel-count, which is to be expected.

As for the 1DsMKII, the 1DMKIII will surely provide (pixel-by-pixel or sensel-by-sensel) superior performance in pretty much all aspects, except for total pixel count. The 1DsMKII is capable of beatiful, razor-sharp images, assuming its optics follow. However, when it comes to reproduction of tonal-gradients, color gamut, high-ISO performance, long-time exposures, the 1DMKIII should clearly come ahead (it is to be expected, anyway, and should not be of surprise).

One interesting thing, though, would be to re-design the 1Ds and produce it with detacheable grip, weather sealed (as much as you can), and same 16 MPixel count, BUT, this time, with 1D MKIII sensel technology and none of the high-speed performance, at a price slightly lower that of 1D MKIII (close 1D MKII-N price, indeed), but with no high-speed pipeline (up to 4fps), no live-preview, no dual media-slots, no dual-RISC processor for focusing, etc. Just the fundamentals required by this segment of professionals.

That would be a value proposition that would capture the interest of FF-followers and studio professionals, and will blow Nikon out of the water, in this particular segment, with a world-class 16 Mpixel FF sensor.

In any case, the former 1DsMarkII continues to be an excellent camera, capable of reproducing amazing pictures, with amazing detail, especially for those that require larger files. I do not require such large files, and I rather require more-efficient, higher per-sensel quality, and smaller files, for the type of work and photography we do.

Just my 0.02

 

[Asher Kelman]

Paul:

One interesting thing, though, would be to re-design the 1Ds and produce it with detacheable grip, weather sealed (as much as you can), and same 16 MPixel count, BUT, this time, with 1D MKIII sensel technology and none of the high-speed performance, at a price slightly lower that of 1D MKIII (close 1D MKII-N price, indeed), but with no high-speed pipeline (up to 4fps), no live-preview, no dual media-slots, no dual-RISC processor for focusing, etc. Just the fundamentals required by this segment of professionals.

That would be a value proposition that would capture the interest of FF-followers and studio professionals, and will blow Nikon out of the water, in this particular segment, with a world-class 16 Mpixel FF sensor.

So this would be a dream version of the 5D that could focus really well! I'd buy that at 16MP!

1DsMarkII continues ...reproduc(es) amazing pictures, with amazing detail, especially for those that require larger files. [I do not require such large files, andB] I rather require more-efficient, higher per-sensel quality, and smaller files, for the type of work and photography we do.
[/B]

and which cameras do you suggest for this as examples and what you have chosen?

Asher

 

[Ferenc Harmat]

Correct, Asher...

So this would be a dream version of the 5D that could focus really well! I'd buy that at 16MP!

In fact, the idea would to completely eliminate the 5D itself, and replace it with the 1Ds Mark III, which would be a higher-megapixel, modular camera system that well lend itslelf for more studio/landscaping or anything that can be exposure/captured without having to worry about the next "instant".

It should be priced as original 5D or just a notch lower (not that much), and would probably be the best overall FF solution, with a higher sales potential, and on a class of itself, which is the same positioning principle of 1D-class products.

and which cameras do you suggest for this as examples and what you have chosen? Asher

It would pretty much depend on your application, but if we are talking about efficiency and quality packed-in-the-file, then:

1. For the smallest and most "pocketable" performance, with best possible detail in the ISO100-400 band, the Fuji F30 and its latest F31d (I believe) version. Nothing matches it.
2. For starters or amateurs on-a-budget, many would be surprised at how well the latest Pentax K100D performs, especially at high-ISOs, with excellent detail and acuity albeit more false-color rendition close to Nyquist. Next on this list would be Nikon D40, with 6mp, capable of delivering very good ISO-performance and detail, too.
3. In the mid range, the best overall contenders are Pentax K10D, Canon 30D (which, with "only" 8.2 MP already puts better high ISOs performance and per-pixel-detail than most 10 MP solutions), and the rest of the Nikons here do not seem to perform as well, gram-per-gram as the smallish D40. I discard the Fuji S5 because of its huge raw files, and somewhat questionable quantum efficiency potential, due to the break-down of its sensels into two parts, which I am not sure how much they help at high sensitivities/speeds. Another runner-up with impressive potential (per-pixel) is the Sigma SD14. The question here is how much this technology has matured, beyond simple megapixel increases. For instance, how well unwanted "spectrum" or "light colors" are handled once they go pass their target layer, and how much of it lands on the last layer, on what kind of signal-to-noise ratio can you really pull off from this sensor (especially chroma channels) at higher ISOs.
4. In the high / professional range, the 1D MarkII-N takes the crown, simply because it beats any APS-C 8.2 Mpixel implementation, from Canon or any manufacturer, and it is dream camera/body and a pleasure to work with, too. As an example, just go to www.imaging-resource.com, and grab any ISO100 to IS01600 samples of their "organic", multi-purpose target, and compare inch-per-inch, on the screen with, say, Nikon D200 (10Mp), Nikon D2xs (12Mp), etc. Watch detail rendition on convoluted, textured areas (such as the multi-color napkins or hanging cords or white/bright fabric, on the right). A real eye-opener, indeed: on a class on its own, all the way up to ISO1600... up until the arrival of the 1D Mark III, that is...

I come all the way from Olympus C700UZ, then through Fuji/Nikon and then into Canon, finally. My first Canon was the 1D MarkII, then the N, and now the 1D Mark III is on its way (once it becomes available).

Just some food for thought, here...

 

[Paul Bestwick]

OK......now, does the MKIII capture in 14bit & export as 14bit, 12 bit or interpolate to 16 bit.
A major consideration in determining image quality

Paul

 

[Asher Kelman]

AFAIK, it captures in 14 BIT and processes in 14 BIT and exports in 16 BIT. The capture we'd better check in the white paper referenced above.

Asher

 

[Peter Galbavy]

Reading the white paper and the detail on the new focus sensors, I would have thought for a vertically shooting PJ a mode to rotate between the "top" (when in portrait orientation) group of cross sensors, and not just "outer" and "inner" ?

I currently use the centre + rotate settings on my 1DMkII as per Chuck's white paper and find that I simply just dial past the lower (again, when in potrait orientation) points 100% of the time.

Anyone else see the logic in that ?

 

[John Sheehy]

OK......now, does the MKIII capture in 14bit & export as 14bit, 12 bit or interpolate to 16 bit.
A major consideration in determining image quality

The RAW format has 14 bits. I have two ISO 100 RAW files from the mkIII on my other computer.

 

[John Sheehy]

The RAW format has 14 bits. I have two ISO 100 RAW files from the mkIII on my other computer.

All 14 bits are used; no gaps in the RAW histogram (although there is a slight periodicity to the histogram, suggesting that some lines might be scaled arithmetically, perhaps to match the multi-line readouts).

Read noise is about the same at ISO 100 as it was with the mkII (1.22 ADU vs 1.27, 12-bit); not really any increase in DR at ISO 100, as DR is limited far more by read noise than shot noise at low ISOs in DSLRs. Perhaps the "increased DR" only applies to high-ISO.

 

[Ferenc Harmat]

It will all depend on how the cam performs at ISO50, and what effective/real sensitivity it offers at such speed. (ISO75-80 on the 1D MarkIIN, probably the same on Non-N).

On the 1D MarkIIN (and presumably the non-N), performance is simply superb, at this sensitivity (clearly better than ISO, especially when pushing the shadows).

Once mine arrives (probably May or June), I will be coming back here and post some results.

Until then, stay tuned!

 

[John Sheehy]

It will all depend on how the cam performs at ISO50, and what effective/real sensitivity it offers at such speed. (ISO75-80 on the 1D MarkIIN, probably the same on Non-N).

If it's capable of a full stop more exposure than ISO 100, that would be a big plus; an ISO 50 that is uncompromised. The current implementation is a bit dodgy; you can get 2/3 stop more real absolute exposure at ISO 50, but the read noise floor relative to saturation is 1/3 stop higher than ISO 100.

On the 1D MarkIIN (and presumably the non-N), performance is simply superb, at this sensitivity (clearly better than ISO, especially when pushing the shadows).

I guess that depends on how you define shadows ... bright and medium shadows can enjoy the lower shot noise, but the deepest shadows, relative to saturation, are noisier at ISO 50 because read noise is not improved over ISO 100.

 

[John Sheehy]

What's the point of having 14 bits for people who haven't the faintest idea of what that means? ;-)

You cannot edit an 8 bit JPEG and then think you can make any comments at all on the 14 bits RAW original.

I have, however, converted two 14-bit mkIII RAWs with my own code, and even with 6x the JPEG compression of the portrait.jpg samples, no such lines are visible with the same treatment.

I know what's JPEG, and what's RAW. I've spent hundreds of hours studying these things, and anyone familiar with the artifacts knows that those lines in the portrait are not from JPEGs, or from 8 bits.

 

[Ferenc Harmat]

Whenever you wish to analyze a pair of exposure-controlled samples (ISO50 and ISO100), from the N, .CR2, let me know.

I will send them to you via www.yousendit.com, if you leave me your desired email address on a private message.

In this way, you can run your own analysis, with your own tools, and share those findings. That would be interesting.

Your call.

 

[Ray West]

Hi John,

I couldn't pm you, so showing my ignorance, or the poor quality of my eyesight/monitor, but can you point out to me the banding you mentioned in the 'hair shot' posted? I'm in no way saying its not there, but I can't see it amidst all the noise. I've seen other banding issues in other posts, but I'm not sure what I'm looking for in this one.

Best wishes,

Ray

 

[Bart_van_der_Wolf]

... can you point out to me the banding you mentioned in the 'hair shot' posted?

Try scolling the screen display vertically (just stating the obvious direction), and you'll see a hint of vertical patterns/striping. It seems doubtfull whether they will show on regular images.

Bart

 

[Ray West]

Hi Bart,

thanks, are they pretty close together i.e. if the image is 100mm wide, then bands about 1mm apart, or something else? A bit like those 'stareos' of a few years back, going cross eyed trying to see the dolphins...

Best wishes,

Ray

 

[John Sheehy]

Try scolling the screen display vertically (just stating the obvious direction), and you'll see a hint of vertical patterns/striping. It seems doubtfull whether they will show on regular images.

A camera like this should be useful for more than pictures shot literally. What is all the point in the hoopla about low noise, high dynamic range, etc, if it isn't possible to take advantage of it?

The lines in portrait.jpg are very easily seen by using Photoshop's "Shadow/Highlights" tool with the default values for shadows (50/50/30). I can see lines in the highlight areas of the face as well. It is most apparent in areas where there are transitions between different tonal areas, as if there were some kind of sag in the amplifiers (I believe that the problem with some early Nikon D200s was the same).

Another way to emphasize the line effect is to run a motion blur of 40 pixels or so in the direction of the purported lines, that leaves the lines and hides the random noise that may mask the lines.

Fortunately, this seems to be a problem limited to a specific specimen. The RAW files I downloaded from IR look very good. Only thing about them is that they are aliased quite a bit - the AA filter is either very weak or non-existent; there is color moire on all high-contrast edges (just like Leica M8 RAW files). This means that my simple demosaicing code doesn't work well - the image needs to be chromatically unsharpened a bit in the conversion process. Cameras with strong AA filters can get away with simpler interpolation.

After seeing the couple of ISO 100 RAWs, I can only say that the JPG samples from Canon were botched; 700KB JPEGs I make from the RAWs are far superior in terms of noise, posterization, and any other artifacting (other than the moire), than the 4.5MB Canon samples.

 

[John Sheehy]

Hi John,

I couldn't pm you, so showing my ignorance, or the poor quality of my eyesight/monitor, but can you point out to me the banding you mentioned in the 'hair shot' posted? I'm in no way saying its not there, but I can't see it amidst all the noise. I've seen other banding issues in other posts, but I'm not sure what I'm looking for in this one.

These bands are only one pixel wide, and are sort of periodic, so they are easy to miss with a soft monitor, or soft vision. On my LCD, I can see them at 100, bareley, without any PP.

With the PS Levels tool, using the black and white droppers for dark and light areas of the haor to the left of the neck, they come out, and also with the "Shadow/Highlights" tool, with default shadow values (50/50/30).

At 100%, they might be harder to see with a soft CRT, an interpolated LCD, or soft vision.

 

[Ray West]

Hi John/Bart,

I found even just loading it into photoshop showed the bands, (different colour space than bare screen) the motion blur is an interesting use for emphasizing it, but it was visible anyway. In the small navigator window the bands 'clump together' nicely, and also if the image size is set to 28.5%. As you say, John, it shouldn't be there, but I think you also said that it seemed to be only that one image, and their jpeg conversion was in any case not so clever.

Best wishes,

Ray

 

[Ferenc Harmat]

This is very interesting...

The RAW files I downloaded from IR look very good. Only thing about them is that they are aliased quite a bit - the AA filter is either very weak or non-existent; there is color moire on all high-contrast edges (just like Leica M8 RAW files). This means that my simple demosaicing code doesn't work well - the image needs to be chromatically unsharpened a bit in the conversion process.

This could be true, especially now that the cam has a dual-DiGiC "core": some of these processing functions could actually now be implemented in software or silicon, rather than an actual, physical filter. It could be.

However, considering what I have seen from my 1D MarkII files (as well as "N", now), I would not be surprised, at all, if the AA filter has been always soft, indeed). Here is a "torture" file, captured with my older 1D MarkII, which proved to be a nightmare for pretty much any RAW-to-RGB converter except C1 or RSP:

http://www.jirvana.com/raw_large/1d_mkII/1dmk2_microdetailraw.zip

Check it out, and see if there is any similar behaviour to what you are generally seeing (on the Aliasing department) on the 1D Mark III.

 

[John Sheehy]

Check it out, and see if there is any similar behaviour to what you are generally seeing (on the Aliasing department) on the 1D Mark III.

Yes, lots of strong color moire at the right sife of Koren's chart (and a little on the door to the right of it).

 

[Bart_van_der_Wolf]

A camera like this should be useful for more than pictures shot literally. What is all the point in the hoopla about low noise, high dynamic range, etc, if it isn't possible to take advantage of it?

Quite true, but it's my opinion that;
a. The cameras are preproduction models.
b. Canon's examples are notoriously non-representative of what's possible.
c. JPEGs are in camera processed samples (adding tonemapping to a and b), and thus it is hard to judge the real/Raw behavior of the sensor array. However, I do agree that they should produce nice JPEG workflow examples (see b).
d. On screen viewing at 100% zoom, is about 3x larger than in print, and printing will blur a bit.

The lines in portrait.jpg are very easily seen by using Photoshop's "Shadow/Highlights" tool with the default values for shadows (50/50/30).

Those settings, to me anyway, are excessive defaults and will show mostly posterization artifacts.

I can see lines in the highlight areas of the face as well. It is most apparent in areas where there are transitions between different tonal areas, as if there were some kind of sag in the amplifiers (I believe that the problem with some early Nikon D200s was the same).

I don't see many artifacts (on my calibrated LaCie 321), except for increased posterization after S/H, but if you do it may be due to your screen, or a proof of a raw processing error introduced after capture. Highlights have the lowest shot noise of an image, so artifacts should then originate from Raw processing.

Fortunately, this seems to be a problem limited to a specific specimen. The RAW files I downloaded from IR look very good. Only thing about them is that they are aliased quite a bit - the AA filter is either very weak or non-existent;

I tried analyzing the resolution, noise, and approximate gamma of the JPEG images with Imatest. The limitations of JPEG, and the unknown post-processing, make reaching objective conclusions a tricky exercise. However, the AA-filter seems fine (low modulation beyond Nyquist) if not a bit strong. There is a semi-systematic difference between horizontal and vertical resolution (the EXIF data also reveals a tiny difference in sensel pitch, but that obviously is systematic). There may be other causes for the resolution differences, but I don't know all details of their testing setup and safeguards/double-checks. Maybe it's due to the Rawprocessing, maybe there is a tripod-vibration/mirror-slap issue.

After seeing the couple of ISO 100 RAWs, I can only say that the JPG samples from Canon were botched; 700KB JPEGs I make from the RAWs are far superior in terms of noise, posterization, and any other artifacting (other than the moire), than the 4.5MB Canon samples.

Raw processing makes a lot of difference, also on the choices how the various Raw Processors handle false-color aliasing. The only solid test is based on Raw data, linear gamma, raw conversions (preferably several that don't add sharpening artifacts).

Bart

 

[Ferenc Harmat]

Well, it was all a myth, then...

Yes, lots of strong color moire at the right sife of Koren's chart (and a little on the door to the right of it).

...For ages and ages, I heard the "soft AA" fantasy abou the 1D MKII's sensor, and for ages and ages I could see that it was just that: a malformed, street-brewed rumor that could not be evidenced even with the most elemental test, such as the above Koren's chart.

It is with this chart that I noticed how ACR improved along the way, as it outputs rendition of high-frequency detail that it otherwise messed up completely, in past implementations. It is not only the false-color rendition, but the horrible maze/artifacts that it used to show on those areas. By then, C1 was already miles ahead of ACR, and, today, ACR does a much, much better job at rendering out-of-band frequencies by using in-band lower-frequencies as proxy, thus allowing to partially reconstruct what is at or slightly beyond Nyquist in a much more pleasant and "life-like" way.

I will re-test this target, once my 1D MKIII arrives, and we will see how it fares.

Stay tuned...

 

[John Sheehy]

...For ages and ages, I heard the "soft AA" fantasy abou the 1D MKII's sensor, and for ages and ages I could see that it was just that: a malformed, street-brewed rumor that could not be evidenced even with the most elemental test, such as the above Koren's chart.

I've read reports of other people that compared the mkII with the mkIIN, and found that the RAW data was similar, even though at one point they had quite different reputations of AA strength.

It is with this chart that I noticed how ACR improved along the way, as it outputs rendition of high-frequency detail that it otherwise messed up completely, in past implementations. It is not only the false-color rendition, but the horrible maze/artifacts that it used to show on those areas. By then, C1 was already miles ahead of ACR, and, today, ACR does a much, much better job at rendering out-of-band frequencies by using in-band lower-frequencies as proxy, thus allowing to partially reconstruct what is at or slightly beyond Nyquist in a much more pleasant and "life-like" way.

Yes, AA has made some good strides in its lifetime. One thing I noticed was that as of the 3.6 beta, banding noises were quite reduced; at one time, I could reduce the banding myself much more easily, but I noticed that 3.6 gave me some good competition. One area in which ACR still is behind, though, is in uneven amplification of lines - it still makes a checkerboard pattern in cameras where the amplification of lines of pixels are not calibrated properly. DPP seems to recognize this and compensate. DPP does not pay any attention to line-based read noise offsets, though.

Why can't there be one converter that recognizes all the most common digital artifacts?

 

[Ferenc Harmat]

In all fairness...

Why can't there be one converter that recognizes all the most common digital artifacts?

There has been substantial progress on the ACR front, although Canon's EVU/ZB/RIT, C1 and RSP have been able to do a commendable job with pretty much most impurities, without negatively impacting spatial performance.

However, a question remains with their ability to handle analog-rooted issues: I wonder if such converter refinements could be used (by manufacturers) as means of "hiding" these subtle imperfections, and not really working on them, at their source.

Just some thoughts, although, I would definitely welcome these features as options to be enabled or disabled, by the user.

My 0.02

 

[Sean Shadbolt]

I have to say , the only time I have seen this on my 1dsmk 2 is when generating jpegs from Capture one LE - I haven't tried others as that has become my standard converter as I like the colour so much
(don't get me started on ACR!).
I usually generate Tiffs now and convert to jpeg in Photoshop as it makes the issue disappear.

 

[John Sheehy]

There has been substantial progress on the ACR front, although Canon's EVU/ZB/RIT, C1 and RSP have been able to do a commendable job with pretty much most impurities, without negatively impacting spatial performance.

However, a question remains with their ability to handle analog-rooted issues: I wonder if such converter refinements could be used (by manufacturers) as means of "hiding" these subtle imperfections, and not really working on them, at their source.

Well, random noise hides it if the random noise is higher. That's why you don't see the line-banding in shadows on some other brands of DSLRs that use CCDs, at higher ISOs; they have more random 2D noise, which hides the patterned noises. If you actually extract a map of the banding offsets, however, the CCD cameras can actually have stronger banding.

Just some thoughts, although, I would definitely welcome these features as options to be enabled or disabled, by the user.
My 0.02

Certainly many should be implemented as an option. There is no point in a converter wasting CPU cycles looking for unequal amplification patterns on a camera that doesn't show them, or in shots where they don't matter, but it is actually a very simple process. All the converter really has to do is sample the average green value for each line, and detect a pattern, and then scale the values in the lines (red and blue, too), preferably at a higher bit depth (or the depth is already deeper than necessary, as seems to be the case with the 1DmkIII). The value in being able to turn a feature like that off is that you may have taken an almost perfectly regristrated exposure of actual lines, and the feature would weaken them. Even with the slightest angle, however, the lines should be safe.

There may be some amplifier sag in that "portrait.jpg" camera; it's hard to tell without seeing the RAW. That would probably require some profiling, to see how the amplifier reacts to a real B/W edge, and base corrections upon it. ANd really, if you want maximum quality from any camera's RAW files, the converter should know all of its repeatable flaws. The converter could ask you to take several shots of a smooth wall or card, completely out of focus, and use their common artifacts for a flat-field adjustment. It could ask you to take a sharp shot of a big black square on a white background, to see if the amplifiers sag. Individual errant pixels could have their unique sensitivities mapped, etc, etc.

I find it rather sad that almost all post-processing of RAW shots is after conversion. Artifacts are so much more distinct in the RAW state; waiting until after conversion to remove them is like dropping a chocolate bar into a pot of soup, and instead of just removing the chocolate as quickly as possible, letting it melt and stirring it in.

If I could run something like Filtermeister from ACR, and run it on the RAW data just after decompression, I'd be a very happy camper.