IQ3 100MP technical camera tests: color cast, mazing artifact, tiling issue, DR etc

[voidshatter]

I have just done some tests today at Teamwork Photo. Due to time constraints I could not perform all tests I wished. I will keep this thread updated as I dig out new findings. Any input is welcome.

Lenses tested:

HR ALPAGON 5.6/23 MM, LB, with center filter
HR ALPAR 4.0/35 MM, LB
HR ALPAGON 4.0/40 MM, SB17

 

[voidshatter]

First of all, I have found that the IQ3 100MP has less color cast issues than the IQ250 does. At least this is true for the uncorrected LCC files. However, there is currently a bug with Capture One v9.0.2: it will cause residual greenish color cast in the LCC correction! You will need to use Capture One v9.0.1 to make LCC correction, but v9.0.1 does not have a profile for the IQ3 100MP yet and will treat it as the IQ250.

 

[voidshatter]

For this scene I have bracketed with the 23HR, for IQ3100 vs IQ380 vs IQ250.

I am sure that I was shooting the IQ3100 in 16-bit mode, which was clearly displayed in screen.

Observations:

a) I am using the latest RawDigger (version 1.2.3). It is perhaps not adequate enough to decode the IQ3100 properly yet. This was also found in the darkframe test shots.
b) When I look into the histogram of the shadow, I have found that the IQ3100 is just 14-bit which is the same as the IQ250 (i.e. there are gaps), while the IQ380 seems to be really 16-bit (with no gaps).
c) If the current version of RawDigger is to be trusted (which I doubt) then the fair exposure compensation for the lowest ISO settings of each back at the same aperture is as below:

IQ3100: 1/10s to retain most highlight details in the sky
IQ380: 1/2s to retain most highlight details in the sky
IQ250: 1/4s to retain most highlight details in the sky

Proof was the calculation of average levels of each channel in the same sky area:

 

[voidshatter]

The following comparison was assuming that the highlight headroom calculations by RawDigger v1.2.3 was correct and they retained the same amount of highlight details. Now we see how the shadow pushes. I have applied the following in Capture One v9.0.1 (which is not officially supporting IQ3100 yet, but I do not see cleaner shadows with Capture One v9.0.2): I have cropped out the IQ3100 and the IQ380 to match the crop sensor of the IQ250. I have disabled noise reduction. I have applied +100 shadow, +100 highlight, then compensated exposure to acquire similar resulting histograms.

Findings:

a) The IQ3100 is not necessarily cleaner here, hence not necessarily more dynamic range! Perhaps something is wrong by relying on the calculations of RawDigger v1.2.3.
b) There can be visible chromatic aberration with the IQ3100 (even with Capture v9.0.2). I am not sure whether this is a software issue or simply a property of the sensor.

 

[voidshatter]

Here I attach the whole frame of the processed pictures so that you could see the tonality of each image. Note that when the 23HR was mounted on the IQ250 it was shifted 6mm, but unshifted on the other two fullframe sensors.

 

[voidshatter]

Okey, it seems that we should trust Capture One v9.0.2 instead of RawDigger v1.2.3. By converting the IIQ files into DNG files with Capture One v9.0.2 then opening the DNG files with RawDigger v1.2.3, I found the following:

Capture One v9.0.2 can indeed squeeze out one more extra stop of highlight headroom out of the IQ3100! The "correct" exposure compensation (for the same aperture) should now be:

IQ3100: 1/4s ISO50
IQ380: 1/2s ISO50
IQ250: 1/4s ISO100

 

[voidshatter]

Now by trusting Capture One v9.0.2 for decoding highlight details out of the "16-bit" IQ3100, with the new exposure compensations, we have the following shadow recovery:

The IQ3100 CMOS indeed has cleaner shadow than the IQ380 CCD does!

 

[voidshatter]

And now we look into the DNG files again and we can find that the IQ3100 DNG file is indeed "true 16-bit" without the gaps of levels in shadow. It appears that Capture One v9.0.2 can either correctly decode the IQ3100 RAW file into 16-bit or properly cook the IQ3100 RAW file into 16-bit. The IQ250 remains unlucky here as 14-bit.

 

[tjv]

As you've said re. the chromatic aberrations, and maybe this is related to that, the 100mpx file shows some strange colour that the other two don't in what should be flat areas of tone. Even with your second lot of images, I'd consider the results in noise a wash though, especially considering the +100 adjustment to the shadow slider.

 

[dougpeterson]

As I'm sure your dealer let you konw 9.0.2 does not properly support LCC for IQ3 100mp. That will be addressed shortly, and should be ready just in time for DT's comprehensive multi-back multi-lens tech camera test!

 

[vjbelle]

First of all, I have found that the IQ3 100MP has less color cast issues than the IQ250 does. At least this is true for the uncorrected LCC files. However, there is currently a bug with Capture One v9.0.2: it will cause residual greenish color cast in the LCC correction! You will need to use Capture One v9.0.1 to make LCC correction, but v9.0.1 does not have a profile for the IQ3 100MP yet and will treat it as the IQ250.

Yes...... I have never seen that green cast before that I've seen in other posts. Thanks so much for pointing this out. I even did some testing with my Leaf/Credo 50 and shifted my 35mm XL 15mm and still couldn't find that cast. In fact the latest version of C1 actually gave me an extremely nice file...... even at 15mm Rise!! So..... I have greater expectations for the 100MP.

Victor

 

[yaya]

Please note that a fix for the LCC bug is due shortly in a service release...probably a matter of days...

BR

Yair

 

[voidshatter]

Just leave the DR thing for a while. Now moving to the 40HR already! Below is IQ3100 vs IQ380. Here I have applied the following: +25 highlight, +25 shadow, "Pick White Balance Tool" on the same spot, LCC correction (by Capture One v9.0.1), lens correction with 40HR profile and movements.

Because when I was testing this, it suddenly rained, and I had to escape, without shooting the IQ250 at all. (Just one less justification to buy technical cameras for landscape in changing weather conditions!)

Here you go: IQ3100 vs IQ380, 0mm up, 10mm up, 18mm up

 

[voidshatter]

I can't wait to do real pixel peeping! At f/9, the 100 MP is now the new king!! Just check the details on that wire of the CCTV! (You'll need to click on the image below to view it enlarged)

Note that I have ticked "Chromatic Aberration" for lens correction, and this is the same lens unmoved and unshifted.

 

[thrice]

Interesting purple glow in the last IQ3100 but not the IQ380, almost looks like optical purple fringing due to defocus, are both backs shimmed?

 

[voidshatter]

Now moving to mazing artifact. For those of you who are not familiar with mazing artifact you may refer to my old thread about the IQ250.

When does it happen? When you shoot indoors (presumably with narrow band light source conditions), it is mostly visible when you zoom into 100% for pixel peeping in an area where crosstalk is "worst" and demosaicing is having issues.

For the IQ250, look for the blue part of the LCC file. This time for the IQ3100, it has changed a bit: look for the region between the blue cast, the maganta cast and the yellow cast! (as marked below)

I am happy to announce that if you stay within the official image circle of the 23HR then it is free of mazing artifact (as far as I can see). Also keep in mind that for the same part of the picture, you may get mazing artifact if you shift too much (as shown in the example below) so don't be greedy when you shift on a Sony CMOS like this! "Shifting too much + stitching + cropping" could be worse than "shifting just enough + stitching" because if you are not careful then mazing artifact may ruin your work!

+5mm shift of 23HR = mazing artifact about to start
+10mm shift of 23HR = mazing artifact

Given that this is a Digaron-S lens with only 70mm image circle, I don't think mazing artifact is to be worried about.

 

[JeRuFo]

Does the CA look comparable on the 380 without the CA correction on?
Is this a lens profile problem or a back problem?

 

[voidshatter]

Interesting purple glow in the last IQ3100 but not the IQ380, almost looks like optical purple fringing due to defocus, are both backs shimmed?

None of these backs were shimmed. I used the LiveView of the IQ3100 to focus carefully. It's hard to explain why neither the IQ380 nor the IQ250 have that purple fringing, while the IQ3100 does give most details for pixel peeping.

 

[voidshatter]

Does the CA look comparable on the 380 without the CA correction on?
Is this a lens profile problem or a back problem?

If I uncheck CA correction for both then I get the following:

 

[JeRuFo]

If I uncheck CA correction for both then I get the following:

Thanks. That is troubling. Could be the pixel pitch getting into play.