X2D 100C Firmware Update (2.0.0)

[JimKasson]

I actually don’t understand how a JPG histogram is less work - the raw is being read off of the sensor by something, and adding a step to histogram the values between 0 to 2^n-1 bits seems less mathematically challenging than running the JPEG algorithm and then doing the same thing (maybe at a lower resolution), but still. The the EVF/display basically take raw data anyway at a lower res, not JPEG data. So why not histogram it in that domain?

The subsampling etc needs to be done for the EVF display anyway. Building the histogram off the subsampled data seems to be the path of least resistance. But I don't think splitting the image chain and doing a raw histogram would be that big a deal if the product managers asked for it. I think the issue is that they're not asking for it.

 

[doccdiamond]

Yes it did/does, not sure about the later versions....

john

The later versions not…- for whatever reason

 

[hcubell]

Was the glitch with the X2D's default WB being reset to the WB setting of the first file downloaded into Phocus from a card or the SSD, when the camera had been set to AWB, fixed in the new FW?

 

[MGrayson]

Was the glitch with the X2D's default WB being reset to the WB setting of the first file downloaded into Phocus from a card or the SSD, when the camera had been set to AWB, fixed in the new FW?

Nope!

 

[JimKasson]

There's an error in the focus bracketing section of the new manual:


The pixel pitch of the X2D is 3.76 um. I don't know whether the extra small CoC is 1 times the pitch, or 5.3 um. It can't be both.

 

[MGrayson]

There's an error in the focus bracketing section of the new manual:

View attachment 203881
The pixel pitch of the X2D is 3.76 um. I don't know whether the extra small CoC is 1 times the pitch, or 5.3 um. It can't be both.

Those are the numbers from the X1D. Someone forgot to divide by the square root of two!

 

[JimKasson]

Those are the numbers from the X1D. Someone forgot to divide by the square root of two!

Divide which by root two? The pixel pitch? That leaves us with the question, is the extra small CoC 5.3 um or 3.76 um. There are arguments for making it either.

 

[MGrayson]

Divide which by root two? The pixel pitch? That leaves us with the question, is the extra small CoC 5.3 um or 3.76 um. There are arguments for making it either.

Well, we can argue what the CoC should be, but 1xPP is, as you said, off by Sqrt[2] = 5.3/3.76 (approximately, but very close). I'm just guessing that the 50MP -> 100MP change got forgotten somewhere. In the end, it's the results that matter, and everyone can choose the stepsize that works best for them.

I should have said that those numbers are consistent with the X1D, rather than that they originated there.

 

[JimKasson]

I'm just guessing that the 50MP -> 100MP change got forgotten somewhere.

The question is whether it was forgotten by the manual writers or by the firmware writers? Oh, I guess it's apparent that it was forgotten by the manual writers. But did they change the CoC sizes in um when they came up with the new firmware?

 

[JimKasson]

In the end, it's the results that matter, and everyone can choose the stepsize that works best for them.

The testing necessary to determine the step size vs camera setting is not a simple thing. At least I don't know a simple way to do it. Having it in the manual is a big help, if only we knew which formulation is the right one.

 

[MGrayson]

The question is whether it was forgotten by the manual writers or by the firmware writers? Oh, I guess it's apparent that it was forgotten by the manual writers. But did they change the CoC sizes in um when they came up with the new firmware?

The testing necessary to determine the step size vs camera setting is not a simple thing. At least I don't know a simple way to do it. Having it in the manual is a big help, if only we knew which formulation is the right one.

Good question and good point. I imagine that one could do a test with, say, 100 steps and measure the focal plane shift. I might try it, but my experimental standards of accuracy are laughable.

 

[jng]

Good question and good point. I imagine that one could do a test with, say, 100 steps and measure the focal plane shift. I might try it, but my experimental standards of accuracy are laughable.

Matt.

If you still have your X1D, you could try 100 steps with the X1D and X2D, each set at the same interval (e.g., "medium") and see whether the two cameras cover the same distance (or not). A perfect experiment for a theoretician to carry out on a Sunday afternoon!

John

 

[buildbot]

Well, we can argue what the CoC should be, but 1xPP is, as you said, off by Sqrt[2] = 5.3/3.76 (approximately, but very close). I'm just guessing that the 50MP -> 100MP change got forgotten somewhere. In the end, it's the results that matter, and everyone can choose the stepsize that works best for them.

I should have said that those numbers are consistent with the X1D, rather than that they originated there.

Isn’t the sensor 3.76um pixel pitch? Which makes sense as the area resolution doubled, so each dimension decreases pixel pitch by sqrt(2).

 

[MGrayson]

Isn’t the sensor 3.76um pixel pitch? Which makes sense as the area resolution doubled, so each dimension decreases pixel pitch by sqrt(2).

Right. But the chart numbers didn’t change. We think…

 

[MGrayson]

Matt.

If you still have your X1D, you could try 100 steps with the X1D and X2D, each set at the same interval (e.g., "medium") and see whether the two cameras cover the same distance (or not). A perfect experiment for a theoretician to carry out on a Sunday afternoon!

John

Yes, I did 19 steps because I'm an idiot and I took 20 shots. And I had to use the same bracket on both cameras because different brackets put the body in different places.
Both series started focused on the 8" mark. Steps were medium. Thanks to trigonometry, all the actual numbers here are meaningless. It's their ratios we care about. I'll leave actual distance measurements to the pros.

X1D XCD 120/3.5 focused at 2.3 feet (Disto D5, so we can believe SOMETHING here). Here's the 20th image.


I read the X1D final image as focused at 8 15/16", or an advancement of 0.9375". The 9" mark and the 8 7/8" marks look about the same OOF. Maybe. ¯\_(ツ)_/¯ . If the X2D advancement were scaled by the pixel pitch, then we'd expect the last frame to focus at 0.9375 * 3.76 / 5.3 = 0.662913 (false precision R us). Yes, we are deciding in advance what result we want to see and then we’ll squint at the data and believe we’re right. That’s the scientific method as I learned it .

Here's the 20th image from the X2D with the 0.663" advancement marked by the arrow.

Looks good to me. It might be slightly too far. The 11/16" mark (just past the arrow) is slightly sharper than the 9/16" mark (2 ticks closer to us), so true focus is slightly further than 5/8", and the arrow lies in that gap.

Sorry, John, I am not repeating this experiment 5 times.

Matt

 

[jng]

Yes, I did 19 steps because I'm an idiot and I took 20 shots. And I had to use the same bracket on both cameras because different brackets put the body in different places.
Both series started focused on the 8" mark. Steps were medium. Thanks to trigonometry, all the actual numbers here are meaningless. It's their ratios we care about. I'll leave actual distance measurements to the pros.

X1D XCD 120/3.5 focused at 2.3 feet (Disto D5, so we can believe SOMETHING here). Here's the 20th image.


I read the X1D final image as focused at 8 15/16", or an advancement of 0.9375". If the X2D advancement were scaled by the pixel pitch, then we'd expect the last frame to focus at 0.9375 * 3.76 / 5.3 = 0.662913 (false precision R us).

Here's the 20th image from the X2D with the 0.663" advancement marked by the arrow.


Looks good to me. It might be slightly too far. The 11/16" mark (just past the arrow) is slightly sharper than the 9/16" mark (2 ticks closer to us), so true focus is slightly further than 5/8", and the arrow lies in that gap.

Sorry, John, I am not repeating this experiment 5 times.

Matt

Wow, be careful or you'll risk having your theorist's license revoked!

So, do you/we conclude that the firmware is operating according to actual pixel pitch and the X2D owners manual was simply not updated? Just asking for a friend...

John

P.S. Not to conflate Airy discs and circles of confusion, but it seems to me that there may be little actual benefit in using steps smaller than "medium" (i.e., 2 x pixel pitch) given Nyquist limits? Help, where's a theorist when you need one?

 

[MGrayson]

Wow, be careful or you'll risk having your theorist's license revoked!

So, do you/we conclude that the firmware is operating according to actual pixel pitch and the X2D owners manual was simply not updated? Just asking for a friend...

John

P.S. Not to conflate Airy discs and circles of confusion, but it seems to me that there may be little actual benefit in using steps smaller than "medium" (i.e., 2 x pixel pitch) given Nyquist limits? Help, where's a theorist when you need one?

I can't interpret the chart posted above, but yes, the firmware *is* taking the pixel pitch into account.

I'm not going to touch the theoretical issues of CoC. Among other things, there's the resolution lost due to de-mosaicing. Hopefully, @JimKasson will tell us something we can believe about the optimal CoC of a digital sensor.

As for a "practical" test, I'm going to use the old 1980's CPU silicon wafer target. It has detail much finer than anything except expensive targets that go down to optical wavelengths, and we can see what step size leaves detectable loss of detail in the gaps. More modern wafers have all their detail at UV or X-Ray wavelengths.

Next time on "Amateur Scientist!"

 

[JimKasson]

Not to conflate Airy discs and circles of confusion, but it seems to me that there may be little actual benefit in using steps smaller than "medium" (i.e., 2 x pixel pitch) given Nyquist limits? Help, where's a theorist when you need one?

If you are using focus bracketing to get the sharpest single image, rather than stacking, CoCs smaller than the pixel pitch are highly desirable.

 

[JimKasson]

I'm not going to touch the theoretical issues of CoC. Among other things, there's the resolution lost due to de-mosaicing. Hopefully, @JimKasson will tell us something we can believe about the optimal CoC of a digital sensor.

See this for a partial answer:

How Your Camera’s Focus Bracketing System Works

For many years, photographers have had focus-stacking programs like Helicon Focus and Zerene Stacker. You can now get the capability in some general-purpose image editors like Lightroom and Capture One Pro. Focus stacking software takes as input a series of images of the same subject with the...

www.lensrentals.com

 

[JimKasson]

As for a "practical" test, I'm going to use the old 1980's CPU silicon wafer target.

Oooh. I'd love to have one of those.