GFX microlens discussion at Kasson

[hcubell]

It's perfectly fine to discuss things I said in my blog posts. That's different from saying I said something which I don't remember saying, that seems at odds with my other writings at the time, and for which you have no documentation that I said it.

Jim

Your blog about this issue was just a few days ago, and yet you denied today making any statements to the effect that the smaller microlenses in the GFX were leading to the prevalence of "oversharpened" GFX images that you have been seeing on line. The words were right there for all to see. If you have difficulty acknowledging in a candid way what you said just a few days ago (for reasons that completely escape me), how can you be so sure of what you said a few years ago?

 

[JimKasson]

Your blog about this issue was just a few days ago, and yet you denied today making any statements to the effect that the smaller microlenses in the GFX were leading to the prevalence of "oversharpened" GFX images that you have been seeing on line. The words were right there for all to see. If you have difficulty acknowledging in a candid way what you said just a few days ago (for reasons that completely escape me), how can you be so sure of what you said a few years ago?

My blog post stands:

https://blog.kasson.com/gfx-50s/on-microlens-size-in-the-gfx-100-and-gfx-50r-s/

The combination of the small microlenses and overly aggressive capture sharpening is a bad one, and w=the amount of sharpening that is normal for large microlenses is too much for small ones. The small microlenses themselves are more subject to aliasing. I have become more negative about the small microlenses over time, especially with the GFX 100 in hand.

But I have always acknowledged that the small microlenses have downsides. They are sharper, but the aliasing that comes along with that is a problem that seems more significant to me now that I can see a camera with roughly the same microlens size that has 1.4 times the resolution, so the same sharpness, but less aliasing.

Jim

 

[JimKasson]

Your blog about this issue was just a few days ago, and yet you denied today making any statements to the effect that the smaller microlenses in the GFX were leading to the prevalence of "oversharpened" GFX images that you have been seeing on line.

If you look carefully, you'll see that that's not exactly what I said:

"I am not saying that the small microlenses are a problem because they encourage oversharpened images. I'm saying that they are a problem because of aliasing. That's a different thing entirely. The oversharpening occurs, I believe, because Lr and C1 did not adjust their default sharpening in order to take into account the small microlenses, leading to oversharpened images at the default sharpness settings. You can't blame that on the camera. "

 

[dchew]

Your blog about this issue was just a few days ago, and yet you denied today making any statements to the effect that the smaller microlenses in the GFX were leading to the prevalence of "oversharpened" GFX images that you have been seeing on line. The words were right there for all to see. If you have difficulty acknowledging in a candid way what you said just a few days ago (for reasons that completely escape me), how can you be so sure of what you said a few years ago?

Howard,
I'm willing to bet neither of you can be so sure of what you said a few years ago or what was said to you. My wife tells me I can't remember what I said 5 minutes ago. Please, for the rest of us, give it a rest. You both are great contributors here. This microlens issue ain't worth it.

Dave

 

[Shashin]

I was wondering what you guys think of the thoughts expressed in the following article. I believe Lloyd Chambers also wrote something similar.


https://blog.kasson.com/gfx-50s/on-microlens-size-in-the-gfx-100-and-gfx-50r-s/

https://diglloyd.com/blog/2019/20190818_0900-FujifilmGFX100-vs-FujifilmGFX50S.html

If as Kasson suggested, the issue with the GFX50 is due to the microlenses on the sensor, is the same argument also applicable to the X1D sensor, since they are essentially the same. I haven't shot with GFX50, but I don't think I am seeing the crunchiness or oversharpening that is being discussed. Thoughts, anyone?

I would simply test with a print. The information presented in the blog appear to be 100% monitor views, which don't reflect a normal viewing conditions. They also don't take a systemic view of the process to the final image, which can actually result in a perception of softness.

Just as you can sharpen an image, you can also soften it. Is your perceived crunchiness from the sensor or the RAW processor? Look at the sharpening defaults and reduce them.

 

[P. Chong]

Jim (Kasson),

So if we switch off sharpening on C1 or LR or ACR during raw conversion, the oversharpening will go away, though we still have to deal with aliasing? Am I right that the workaround is to stop the lens down?

 

[Paul2660]

I would not turn off all sharpening in C! or LR for a raw conversion, as you are still interpolating colors as with any raw conversion, if you capture R, you interpolate Green and Red, thus softness will be induced.

I have over 10K of images from the 50s, I have yet to find anything wrong with them due to over sharpness. It's a personal preference, I realize but I think I understand the effects of over sharpening.

If any raw conversion uses a special meta lens read from the GFX50s, it might be ACR/LR, as the images due tend to show a tendency to be over sharp, and you have to watch for haloing. I don't shoot subject matter that has ever shown aliasing in landscape subject matter with the 50s 3100 or 4150.

What a lot of people miss is that that what you see on the web can be totally miss represented. Different monitor, different browser, and more important, what tool was used to upload the image. Many, myself included use flickr as one such tool, and for sure it adds sharpening to all my images. You can't turn it off, and if you present a well prepared image for print to flickr, it tends to get an over sharpened look. I personally don't have time to soften images just for web uses. Personally the vast majority of people viewing your work are on a phone or tablet neither of which can give a really good representation of your work, especially at 72 dpi, which is the web standard.

It's no different for printing, as an image to printed on canvas requires a different level of sharpening vs an image on paper. And paper matters also, glossy vs matte and ink sets also. I don't think there is one set level of sharpening that works

I still rely on 100% pixel view in CC or LR, and sharpen accordingly. I use the defaults for sharpening in C1 and less than the defaults in LR/ACR.

I am more concerned about the color profile and noise characteristics of the raw conversion than if it's over sharpened, as I can easily correct for over sharpening, but correcting noise and or color is more time consuming.

Sometimes I find that it's a bit easy to get off track

The world is not blurred and most object have pretty amazing details, both man made and natural and producing an image that displays those details is my goal.

Paul C

 

[JimKasson]

Jim (Kasson),

So if we switch off sharpening on C1 or LR or ACR during raw conversion, the oversharpening will go away, though we still have to deal with aliasing? Am I right that the workaround is to stop the lens down?

Right on all counts, but I suggest just backing down on the sharpening in the raw converter, since turning it off entirely produces soft files. The microlenses on the GFX 50x are small, but they are not point samplers.

Jim

 

[JimKasson]

I would simply test with a print. The information presented in the blog appear to be 100% monitor views, which don't reflect a normal viewing conditions. They also don't take a systemic view of the process to the final image, which can actually result in a perception of softness.

I do test prints a lot. But they have a huge disadvantage for a blogger like myself: you can't show other people what they look like. In my informal tests, I do provide a recipe for the viewer to simulate looking at a print, but neither do I show the softening effect of a real printer using real paper, nor do I sharpen to compensate for that effect.

Just as you can sharpen an image, you can also soften it.

This is true. You can sharpen an image, at a cost of SNR and spatial effects. You can soften an image and pick up SNR, but you may also have spatial effects. (Ask me to explain about the relationship of the space and frequency domains if that's not clear to you.) That in general argues for smaller microlenses.

But here's the catch: If there's aliasing, you can't get rid of it by sharpening or softening. And small microlenses will create more aliasing.

Is your perceived crunchiness from the sensor or the RAW processor? Look at the sharpening defaults and reduce them.

The effect is observable in the four raw planes, but only as a difference between the GFX 50x and GFX 100 files, not as "crunchiness", since you're looking at raw files with no sharpening at all.

I do reduce the sharpening from the overaggressive Lr/ACR results. But it appears that not all people do that, from what I'm seeing on the web. And, as I said before, reducing the sharpening won't help if information is already aliased.

Jim

 

[tsjanik]

Interesting thread. I had not followed the microlens discussions since I am a recent owner of the 50s. I did notice on first use that the 32-64 seemed remarkably sharp, perhaps suspiciously so. When I found that Fuji had lens corrections baked into the RAW files I wondered if there was some file sharpening applied. This discussion motivated me to try a simple test today. Three heavily cropped images from the garden (actual pixels at 72ppi).
In order the images are:
1. 50s using a Pentax 67 55mm at f/8
2. 645Z using the same 67 55mm
3. 50s using the 32-64 mm @55mm and f/8.
Both the images from the Fuji are slightly sharper (look at the center of the sunflower), and since software is not correcting for the 67 lens, my thought of baked-in sharpening can be dismissed, unless sharpening is blindly applied to all files regardless of lens used, which seems unlikely. Assuming the Fuji and Pentax are using the same basic sensor except for the microlenses, my results are consistent with Jim's suggestion. All images on a tripod, timer and zoomed in live view focus; default setting in ACR. I can't dismiss some focus error, although I had two exposures for each and don't see any evidence of such.

Tom

Addendum: The Pentax 67 55mm (last generation) appears slightly sharper than the Fuji zoom viewed at 400%; not very significant but good news for use with a shift adapter.

 

[Shashin]

I do test prints a lot. But they have a huge disadvantage for a blogger like myself: you can't show other people what they look like. In my informal tests, I do provide a recipe for the viewer to simulate looking at a print, but neither do I show the softening effect of a real printer using real paper, nor do I sharpen to compensate for that effect.



This is true. You can sharpen an image, at a cost of SNR and spatial effects. You can soften an image and pick up SNR, but you may also have spatial effects. (Ask me to explain about the relationship of the space and frequency domains if that's not clear to you.) That in general argues for smaller microlenses.

But here's the catch: If there's aliasing, you can't get rid of it by sharpening or softening. And small microlenses will create more aliasing.



The effect is observable in the four raw planes, but only as a difference between the GFX 50x and GFX 100 files, not as "crunchiness", since you're looking at raw files with no sharpening at all.

I do reduce the sharpening from the overaggressive Lr/ACR results. But it appears that not all people do that, from what I'm seeing on the web. And, as I said before, reducing the sharpening won't help if information is already aliased.

Jim

And given the pixel resolution of those sensors under normal shooting conditions with normal subjects, what are the chances that aliasing will be perceptible in a print? Every digital camera will alias. I am sure I have aliasing in my images somewhere, but I have yet to be able to identify it. Ultimately, the goal of cameras in the context of the photographers here is to produce a pleasing image. While the rigor of testing targets can be used to identify issues, the significance of those issues still need to be put into context of the process the systems will be used under, including to the final display.

I think if a photographer is concerned about what you have found, they should be systematic and produce prints to see the systemic effect of the microlenses. The print is the best way to put some of these effects into context, especially with such high MP sensors. That was simply my recommendation.

 

[mristuccia]

Interesting discussion.
Regarding the image sizing for web displaying I just want to add a point: let's not forget that many new device displays don't feature 72DPI anymore, most of them are now above 200DPI.
I personally experience a huge difference in perceived sharpness when seeing the same image on my old full-hd monitor and on my MBP retina display.
Another consequence is that now I don't know what should be the right size of an image for web displaying, the old sizes like 800-1024px long side seem now too soft or too small (depending on the configured scaling factor) on retina displays.

 

[ErikKaffehr]

A bit of perspective...

Hi,

Aliasing was always a part of medium format. Please do remember that essentially all MFD cameras lack OLP (Optical Low Pass) filters, with the exception of the early Kodak backs and the Mamiya ZD that offered an OLP filter as an option.

Combining a good lens with relatively large pixels is a sure recipe for aliasing. But, aliasing yields fake detail instead of true detail. That is the reason it is called aliasing. It converts high frequency detail that the lens transmits into low frequency aliases.

The reason that photographers don't object to aliasing is that it is not very obvious, unless it shows up as color moiré.

Now, for some reason I don't really understand, Fuji claimed to have reduced the pixel aperture on the GFX 50S/R. Any point of the image is smeared out over the pixel aperture, so a small pixel aperture increases contrast at the pixel level, but it will also produce fake detail. So, Fuji essentially choose pixel level contrast over correct reproduction.

Now, we don't know if other vendors have used the same trick. But, wait a moment...

Reducing the pixel aperture means that the pixel collects less light, doesn't it.

Now, let's look at the DPReview test images for the GFX 50/S, GFX 100, Hasselblad X1D and the Phase One XF/IQ 3100MP combo:



We can see that both 50 MP sensors show a large amount of aliasing, while the 100 MP backs are almost aliasing free. Note also that the X1D shows more aliasing than the GFX. That would indicate that either the 90/3.2 lens on the X1D is much sharper or that both use similar pixel aperture.

Reducing pixel aperture significantly would reduce quantum efficiency, leading to longer exposure, but both the X1D and the GFX 50/S use the same exposure f/8 and 1/20s.

Comparing raw exposures for the GFX50S (DSC3041.RAF) and Hasselblad X1D (B0000401.3FR) we can see that there is not a huge difference between G channels:


So both are about 1.4EV under saturation, with presumed saturation at +3EV.

SNR is essentially the square root of the photoons captured.

GFX 50S has SNR = 6821 / 56.9 -> 110
X1D has SNR = 28375 / 236 -> 120

So SNR is pretty close and it seems that the X1D got slightly more exposure (relative saturation). But the differences do not indicate a huge difference in pixel aperture size.

The great unknown here is of course if DPReview uses a variable light source. Their description of the protocols used could be more comprehensive.

My take is really:

• MFD users like aliasing, as it gives an impression of sharpness.
• Fuji with the GFX is obliging by using relatively small pixel apertures.
• The Hasselblad X1D gives similar amount of aliasing as the GFX 50.

Jim's investigation shows that the GFX 50/S has higher MTF at similar cy/mm than the GFX 100. So the GFX 100 obviously has smaller physical pixel aperture than the GFX 50/S. So the GFX 100 has no more acutance than the GFX 50/S, but resolves finer detail.

I always said that both the GFX 50/S and the X1D cried for a higher resolution sensor. Jim's findings show why!

Best regards
Erik

 

[JimKasson]

Re: A bit of perspective...

Hi,

We can see that both 50 MP sensors show a large amount of aliasing, while the 100 MP backs are almost aliasing free. Note also that the X1D shows more aliasing than the GFX. That would indicate that either the 90/3.2 lens on the X1D is much sharper or that both use similar pixel aperture.

Or that the GFX image is misfocused.

 

[ErikKaffehr]

Re: A bit of perspective...

Or that the GFX image is misfocused.

Good point!

Anyway, I think that both systems seem to be able to produce excessive aliasing.

Just to say, I would think that the GFX 50R image has slightly less accurate focus compared to the GFX 50S image.

Best regards
Erik

 

[JimKasson]

Aliasing and printing

And given the pixel resolution of those sensors under normal shooting conditions with normal subjects, what are the chances that aliasing will be perceptible in a print? Every digital camera will alias. I am sure I have aliasing in my images somewhere, but I have yet to be able to identify it. Ultimately, the goal of cameras in the context of the photographers here is to produce a pleasing image. While the rigor of testing targets can be used to identify issues, the significance of those issues still need to be put into context of the process the systems will be used under, including to the final display.

I think if a photographer is concerned about what you have found, they should be systematic and produce prints to see the systemic effect of the microlenses. The print is the best way to put some of these effects into context, especially with such high MP sensors. That was simply my recommendation.

I took a crack at this:

https://blog.kasson.com/gfx-50s/visibility-of-aliasing-gfx-50r-100-aliasing-in-prints/

The question of the desirability of aliasing is a separate issue, but it was clear in the era of 16 and 22 MP MF cameras that some people not only tolerated, but enjoyed, aliasing.

Jim

 

[rdeloe]

I wasn't sure how seriously to treat this issue until I made a picture today that shows the problem vividly. I'm using a Fuji GFX 50R. The lens is a Rodenstock Rodagon-WA 60mm f/4.

The RAW developer used on the RAF plays an interesting role. I tried three RAW developers: Lightroom, Iridient X Transformer and Capture 1 Express for Fuji (exporting as a TIFF that I brought into LR).
* Lightroom showed the least aliasing but the most "mazing".
* IXT and C1 were about the same for mazing and aliasing. Oddly, the aliasing appeared in bands in C1 and IXT, but not in LR.

I then had a look at the JPEG that was created alongside the RAF: no aliasing is evident, and the mazing is minimal. The same is true when you develop the RAF in camera and save it as a TIFF. That's not an ideal solution, but in the cases where the problem is severe, it's a decent work-around.

These are crops to what would be roughly 300%. In order from left to right, Top row: Lightroom, IXT, C1; Bottom row: in-camera JPEG, in-camera TIFF.

Now I'm wondering what it is that the camera is doing with the RAF that these RAW developers are not doing....

Click on this link to see the different versions (goes to my Google drive): https://drive.google.com/open?id=1WBB5pB7XlZEe92U7TqWPu21MBN1hlO_h

 

[rdeloe]

I find interesting how much difference the RAW converter makes. On the same Fuji GFX 50R RAF file, I get dramatically different results using Lightroom 8, Capture 1 Express for Fuji, Iridient X Transformer, and the camera itself.

In an area showing the effects of aliasing in an image I tested yesterday, Lightroom left behind the strongest "mazing" pattern, but the fewest large-scale colour aretfacts. Capture 1 and IXT had less mazing but strong colour artefacts in the form of moire banding. The in-camera JPEG processor left behind a tiny bit of mazing and no colour artefacts. In-camera RAW processing produces a TIFF that (not surprisingly) matches the JPEG relative to colour patterns and mazing.

A pleasant surprise was the new Enhance Details function in Lightroom 8. It did a very good job of cleaning up the colour and mazing artefacts. It's definitely not perfect (can't make a silk purse out of a sow's ear), but it's a vast improvement. Calling it "Enhance Details" is a bit of a misnomer in my opinion. It leaves the rest of the image alone and focuses just on problems areas -- very much like the tools for tidying up chromatic aberrations.

 

[JimKasson]

I find interesting how much difference the RAW converter makes. On the same Fuji GFX 50R RAF file, I get dramatically different results using Lightroom 8, Capture 1 Express for Fuji, Iridient X Transformer, and the camera itself.

In an area showing the effects of aliasing in an image I tested yesterday, Lightroom left behind the strongest "mazing" pattern, but the fewest large-scale colour aretfacts. Capture 1 and IXT had less mazing but strong colour artefacts in the form of moire banding. The in-camera JPEG processor left behind a tiny bit of mazing and no colour artefacts. In-camera RAW processing produces a TIFF that (not surprisingly) matches the JPEG relative to colour patterns and mazing.

A pleasant surprise was the new Enhance Details function in Lightroom 8. It did a very good job of cleaning up the colour and mazing artefacts. It's definitely not perfect (can't make a silk purse out of a sow's ear), but it's a vast improvement. Calling it "Enhance Details" is a bit of a misnomer in my opinion. It leaves the rest of the image alone and focuses just on problems areas -- very much like the tools for tidying up chromatic aberrations.

I can’t argue with any of that.