Fujifilm GFX100 Pixelshift Firmware is out

[Paul2660]

One other note, I have found that in Multi-Shot mode, (3 shot exposure bracket), with a Lexa 1666 128GB card, you quickly buffer out and the camera delays exposure. This is with Performance boost ON. The only card I have found that can keep up is the 2000x Lexar and Sony.

My point being, one pixel shift sequence is 16 shots, so if you fire 2 in quick succession without a fast 2000x card, odds are you will run into the camera buffer issue.

This is with you not tethered.

Paul C

 

[Paul2660]

Frame averaging or dual exposure either one would have been a huge asset. Both may be impossible due to the processor on the camera. Dual exposure for me is still the single most impressive feature of the IQ4 since announce. Especially since it can handle subject movement in most situations.

Paul C

 

[andreleon]

Hi all. I' ve downloaded and pasted the new firmware on the SD card just as I always do . The GfX 100 screen shows" Firmware not found in card slot 1 " (same in card slot 2 ).No success after 3 or 4 attempts. Thanks for your help . Best regards

 

[Paul2660]

Only thing I can thing of, make sure you did not have the previous GFX firmware still on your computer. I did and for some reason this latest and the firmware from 07_20 have the same name. So if you bring in another one with the same name you should see a (1) after the name, if that's there it's won't load correctly.

If you have the name correct, you might try re-formatting card in camera, then drag and drop the firmware onto the card. If you are on a Mac, formatting the card on a Mac, even Mac version of Fat32 (Exfat) might be the problem. I never have much luck if card has been formatted on a Mac, and I drag and drop, so I always format in camera or on a windows machine.

Just a few thoughts.

Paul C

 

[etrump]

While there example is not great, the multi-shot file is pretty impressive photographed both at f8. Both opened with Adobe camera raw without standard open.

GFX Multi Shot Sample

 

[etrump]

Did you format the card on the GFX?

Hi all. I' ve downloaded and pasted the new firmware on the SD card just as I always do . The GfX 100 screen shows" Firmware not found in card slot 1 " (same in card slot 2 ).No success after 3 or 4 attempts. Thanks for your help . Best regards

 

[andreleon]

It's OK Paul and eTrump I had not seen nor read "Caution"when downloading. Thanks a lot . All the best .

 

[gerald.d]

While there example is not great, the multi-shot file is pretty impressive photographed both at f8. Both opened with Adobe camera raw without standard open.

GFX Multi Shot Sample

Hmm.

Not sure I would call that impressive at all.

The multi-shot image shows the exact same combing artifacts that I have experienced with the multi-shot on the Panasonic S1R.

Totally unusable image. I'd venture that the 100MP image upscaled would look better.

Kind regards,


Gerald.

 

[Paul2660]

Yes, quite a bit of combing with the letters against the background, especially the light yellow against green. Hopefully over time Fuji will improve the algorithm involved and the software that uses it. Fuji is pointing to Fine Art Repo for this feature and the aliasing involved will be troublesome possibly.

Paul C

 

[gerald.d]

Yes, quite a bit of combing with the letters against the background, especially the light yellow against green. Hopefully over time Fuji will improve the algorithm involved and the software that uses it. Fuji is pointing to Fine Art Repo for this feature and the aliasing involved will be troublesome possibly.

Paul C

The entire image is riddled with it. Look at the texture on the spine of the book on the left. A total mess.

I'm intrigued as to where this image was taken from - is there a review somewhere discussing it?

Kind regards,

Gerald

 

[gerald.d]

Just to follow up on this, there has been a lot of discussion about the quality and merits of the GFX's pixel shift.

In this YouTube video, the presenter talks about how he originally was not impressed with the system, but having realised that the instructions Fuji provide for capturing a multi-shot image were incorrect, was very impressed with the resultant file once he adapted his workflow.

In the description to the video, he provides a link to the file that he uses in the video to show the differences before and after changing the shooting technique.

Unfortunately, this file still contains the exact same problem that I am seeing everywhere - severe combing artifacts in areas of high spatical frequency that render the file useless for anyone who is actually interested in producing a 400MP file for either archival or high resolution product photography requirements.

The combing artifacts can be clearly seen at 100% view - zoom in on the "S" in "VALUES" (just below and right of center image), and you will see the problem. This is the exact same problem I encountered with the Panasonic S1R. I'm beginning to wonder whether or not this is a processing issue - could the processing software (C1, LR or whatever) be assuming the provided 400MP file requires demosaicing (of course it doesn't), and that is where the problem lies?

To make it even clearer, here is the S, treated in three different ways.

The first is zoomed 800% from that file. Next up, I down-rezzed the image in photoshop to 100MP, and then zoomed it to 1600%. Finally, I re-uprezzed the file in photoshop. I'm not trying to say the uprezzed example is acceptable - it is not - it is just there for comparison purposes.

How anyone can claim the multi-shot files are acceptable is beyond me. I am endeavouring to reach out to Fujifilm to discuss this issue with them further, but if anyone here has contacts and would be happy to connect me, please do drop me a PM.



Kind regards,


Gerald.

 

[buildbot]

Just to follow up on this, there has been a lot of discussion about the quality and merits of the GFX's pixel shift.

In this YouTube video, the presenter talks about how he originally was not impressed with the system, but having realised that the instructions Fuji provide for capturing a multi-shot image were incorrect, was very impressed with the resultant file once he adapted his workflow.

In the description to the video, he provides a link to the file that he uses in the video to show the differences before and after changing the shooting technique.

Unfortunately, this file still contains the exact same problem that I am seeing everywhere - severe combing artifacts in areas of high spatical frequency that render the file useless for anyone who is actually interested in producing a 400MP file for either archival or high resolution product photography requirements.

The combing artifacts can be clearly seen at 100% view - zoom in on the "S" in "VALUES" (just below and right of center image), and you will see the problem. This is the exact same problem I encountered with the Panasonic S1R. I'm beginning to wonder whether or not this is a processing issue - could the processing software (C1, LR or whatever) be assuming the provided 400MP file requires demosaicing (of course it doesn't), and that is where the problem lies?

To make it even clearer, here is the S, treated in three different ways.

The first is zoomed 800% from that file. Next up, I down-rezzed the image in photoshop to 100MP, and then zoomed it to 1600%. Finally, I re-uprezzed the file in photoshop. I'm not trying to say the uprezzed example is acceptable - it is not - it is just there for comparison purposes.

How anyone can claim the multi-shot files are acceptable is beyond me. I am endeavouring to reach out to Fujifilm to discuss this issue with them further, but if anyone here has contacts and would be happy to connect me, please do drop me a PM.

View attachment 179119

Kind regards,


Gerald.

I have to imagine the kind of actuator used in multi shot backs such as the old 528c or Sinar backs have very very precise movement, that are able to accurately step exactly by either a whole pixel or a half pixel in um.

While image stabilizers can control shake really well, accurate shifting may be harder.

 

[marc aurel]

I guess that Fuji is moving the sensor into a kind of "standard position" for pixel shift - an extreme corner of the shifting range. Probably easier to make controlled shifts from a position that can easily be reestablished. When you switch between pixel shift mode and standard mode you can see that the framing shifts. I noticed that different framing when doing compare shots. I had to reframe in Lightroom for my comparison.

Maybe buildbot is right that it is easier with a multishot-back that has no image stabilization. But keep in mind that the pixel pitch on the GFX100 is smaller than those backs.
Im my test I found the improvement quite visible. It shows the print raster on a book spine in my book shelf, taken with the GF63mm lens.
Pixelshift on the left (upscaled by 300% in photoshop), standard image on the reight (upscaled by 600% in photoshop).

Pixelshift is able to resolve the print raster while standard image is not. But I see some artifacts in the pixelshift image too.

 

[marc aurel]

another example with my best lens, the GF 250mm (at f/6.4)
book spines in linen with letters printed on, upscaled by 400% for pixelshift and 800% for standard image

 

[Paul2660]

The issue is clearly visually shown. The image in left besides the combing where letters meet the grey linen show the issue. The difference in the clarity of the linen material also show the merits of the pixel shift as it’s much better.

Obviously an issue that Fuji either missed or figured was acceptable. And thus Odds are will not get fixed anytime soon. They had almost 2 years to bring this feature to market. But I hold some hope since Fuji has a record of fixing issues with firmware updates.

The issue is always shown at 400 % magnification at least in examples I have seen. Has anyone made a print to see how bad the issue shows up?

In pixel shift for me on Pentax K1 it was more important for the ability to capture a cleaner image ie less noise. But with the current Topaz software tools out there, noise reduction is just so easy a process I don’t worry about it like I used to.

As the Fuji option can’t handle any subject movement I guess for me the entire feature is moot.

Paul C

 

[gerald.d]

another example with my best lens, the GF 250mm (at f/6.4)
book spines in linen with letters printed on, upscaled by 400% for pixelshift and 800% for standard image

View attachment 179122

Marc - many thanks for taking the time to investigate this and share your findings.

Cleary, as Paul mentions, we are seeing the same issues in your files as I have observed and commented upon from elsewhere.

It's intriguing that the combing issue seems to be more of a problem where the high spatial frequency changes occur on the Y axis.

My assumption would be that the relative sensor movements for the 16 shots that are taken are the same across both axes.

So why would this lead to more obvious combing on just one dimension?

Just throwing that out there. I don't have an answer, and would welcome any thoughtful insight on exactly what is going on here.

Kind regards,


Gerald.

 

[marc aurel]

could PDAF masking of the pixels have something to do with it?

 

[marc aurel]

extreme enlargement of an area with those artefacts:

 

[f8orbust]

One issue with pixelshifting (PXS) is that it's incredibly susceptible to camera movement between captures, since movement even as small as one pixel is going to have an effect on the final image. In low frequency areas of the image this may not be an issue, but in high frequency areas there's less chance of it going unnoticed. And when you're taking 16 images per capture, there's a lot of time for the camera to move. Because the GFX only has IBIS, and this is being repurposed for PXS, it isn't available to help compensate any camera movement. Systems like the one Olympus use on their m4/3 cameras - which use the sensor for PXS while maintaining IS working in the lens - have, by design, less chance of artifacts appearing as a result of camera movement in the final image.

So in order to get a proper picture of what the GFX100 (or any camera that offers PXS with no IS) is capable of when it comes to PXS, the camera has to be absolutely rock solid during capture. Outside of a lab, I don't know if that's even possible.

 

[marc aurel]

One issue with pixelshifting (PXS) is that it's incredibly susceptible to camera movement between captures, since movement even as small as one pixel is going to have an effect on the final image. In low frequency areas of the image this may not be an issue, but in high frequency areas I can't imagine it could go unnoticed. And when you're taking 16 images per capture, there's a lot of time for the camera to move. Because the GFX only has IBIS, and this is being repurposed for PXS, it isn't available to help compensate any camera movement. Systems like the one Olympus use on their m4/3 cameras - which use the sensor for PXS while maintaining IS working in the lens - have, by design, less chance of artifacts appearing as a result of camera movement in the final image.

So in order to get a proper picture of what the GFX100 (or any camera that offers PXS with no IS) is capable of when it comes to PXS, the camera has to be absolutely rock solid during capture. Outside of a lab, I don't know if that's even possible.

I came to that conclusion too.
My setup with tripod and so on was probably fine. Using a shorter lens than the GF250 did not solve the issue alone. But I took the image on the 5th floor with wooden floors. After I went down to the basement I did better. But not perfect. So controlled environment seems to be essential. Difficult to do that in the field.

I attach the comparison at 200% and 1600%. The one from 5th floor and the GF250 on the left, the one from ground floor and the GF63 on the right. The lighting is different down there. So the image is less contrasty and sharp. But the artefacts are mostly gone.