To be fair to Yunli I think that this is perhaps just naivety vs anything beyond that (and he did fix it). I appreciate the work he does and I'd hate for us to shoo him away. As per Cindy, I think that the message from Bob covered it.
Now if he'd used one of my images ... (just kidding). Btw, Yunli if you want some other tiling problem images from the 260 to illustrate the point I can send you some.
Rosebud ...3 Member(s) thanked for this post2 Member(s) liked this post
a) The shadow of the IQ3100 has been pushed a tiny bit harder than the IQ250. This is probably due to the algorithm implemented for Capture One when you take exposure compensation into account in post processing.
b) I have opened the RAW files by RawDigger 1.2.4. As you can see for the same small region of deep shadow, the standard deviation (noise) is about the same for both the IQ3100 and the IQ250. The IQ380 apparently falls behind a lot, which is consistent with what we see in the real pictures.
c) Then I checked my darkframe test shots by RawDigger 1.2.4 (black level = 0). I have found that the IQ250 loses the scoring (higher standard deviation) due to some higher luminosity in the corners (possibly due to light leakage of the lens cap).
d) If I abuse the high ISO darkframe shots I find the tiling of readout noise for the two fullframe sensors even with Capture One's calibration as shown below (see the tiling of these two fullframe sensors).
e) By manually selecting a sub-region in each darkframe to avoid the tiling partitions I expect to be able to calculate un-skewed standard deviation of readout noise (which presumably affects dynamic range in real-world usage).
Simple words: these two Sony sensors perform quite similarly.
Thanks to Bart's input, I have analyzed the darkframe shots I made. For the IQ3100 case it was shot in 16-bit mode. I used RawDigger 1.2.4 (black level=0) to manually select a sub-region in each darkframe to avoid tiling of readout noise. Then I used the following to calculate the dynamic range:
DR = ln( (65535 - mean(darkframe)) / std(darkframe) ) / ln(2)
It appears that the DR of this new sensor is really great!
1 Member(s) thanked for this post1 Member(s) liked this post
I have just downloaded Capture One v9.0.3. When I try the color cast correction of LCC, WTF??? I might even prefer the v9.0.1 result for this image...
Note: "Pick White Balance Tool" on the same spot (center of the frame)
We've probably beaten this horse enough, but I have a 3100 over the weekend. Lenses I have are 40hr, 60xl, 90hrsw, 150xl.
If anyone wants anything specific let me know. I can compare with an IQ180 but that's it. Max I can shift is 18mm. Sorry Torger, there is snow everywhere so finding you color for a good crosstalk test might be difficult.
I could find a red barn or some evergreen trees if that helps.
Edit: Forgot about a 70hr and 100hr I also have.
Just tried another: IQ3100+40HR 15mm up - still the same red-ish color cast after LCC correction with Capture One v9.0.3.
Capture One v9.0.3 seems to have improved suppression of mazing artifact. Below shows when the 40HR is shifted outside the official image circle what happens when demosaicer attemps to counter crosstalk issues. This is good news - it means Capture One v9.0.3 has improved usable movement range a little bit (despite the red-ish color cast after LCC correction).
It appears that the purple glow has been fixed (or better suppressed) in Capture One v9.0.3 - good news!
1 Member(s) thanked for this post2 Member(s) liked this post
Thanks for posting these Void.
I agree that the 9.0.3 LCC corrections look, well... bad... Colour cast is very pronounced with the 40HR, which pretty much kills this back for me. Not that I could afford one now anyway, but confirms that the Dalsa 60 is the sweet spot for me.
Maybe Phase will also address the faint blotchy banding that happens with shifts on both the 50 and 100 mp chips. They have known about it at least 1 year now for sure on the 50 mp. Just look at the first Alpa tests where you see the blue sky. You can also see it in many of the dealer provided LCC shots.
You can clearly see the issue in any of the test shots that have shifted past 5mm.
It is very faint and jot as hard as microlens ripple. Phase corrects microlens ripple very well currently.
The fact that they have addressed the mazing gives me hope for the banding.
But the current LCC a process does not address it.
I did a simple test this morning using the 3100, IQ180, 40hr and 60xl-with center filter. Dropped the back down 18mm with the back mounted vertically and horizontally. This is not a test designed to push everything to "expose" problems. These are tests for me to see how the back reacts to my normal shooting (A little ETTR).
These are the 3100 images with LCC applied, dust spots and all:
3100-40hr-NoShift (just pointed the camera up a bit)
They look fine to me. I did notice some faint tiling.
If anyone wants the raws from this and/or from the IQ180 to compare, let me know.
Last edited by dchew; 24th January 2016 at 18:12.
davechewphotography.com2 Member(s) thanked for this post2 Member(s) liked this post
Thanks for posting.
Question, I am confused as you are calling it shift, is it not rise? as it appears you did not shift L or R, but moved the back down or up to pull in more sky on top. But I may have this incorrect. The reason I ask, is that I found the banding more problematic on L and R shift, not Rise/Fall movements (in the 50MP chip)
Just looking at the shots on my NEC, the 40 HR-W on the 2nd shot, shows just a bit of red color cast in the upper left corner, at least to me. Not that it would not be correctable in CI
The 40 HR vertical rise, seems less prone to any cast, and just begins to show the IC vignette.
The 60XL to me is overall red cast, but again may be my monitor. The last 60XL shot to me also has just a bit of red on the left side.
Congratulations on the new back.
I did not white balance these nor do anything else with them except apply the LCC, so the red may just be the difference between the lenses. Camera WB was set to Daylight, and the ICC profile is "Phase One IQ3 100MP Flash." Switching to Outdoor Daylight does remove a small bit of red cast.
Oh and one other tiny bit of info: I am just testing the back as a demo. It's not mine! I wouldn't upgrade for image quality. Just doesn't seem very different, and in fact there are some things the 180 does better besides just color cast. For example, specular highlights on this back still show some purple blooming. I don't get that at all on the IQ180.
But OOOhhhh that Live View...
davechewphotography.com1 Member(s) thanked for this post
As for red color cast on the 40hr shot: If I hover in CI in the upper left, RGB ~110/150/200. If I move in to where that second dust spot is just above the skinny cloud, I get ~110/156/204. So you are right; the image is a little brighter moving in but the red doesn't really change.
Honestly I think the jpg has exaggerated the difference. I can't really see much difference on my NEC. But my NEC is long in the tooth (PA241W).
Hi Dave, you are right, I guess it's the same thing.
It just looked like in the first couple of shots, you had the back horizontal and the movement was downward/fall and thus created more sky in the image. The only reason I was curious, is that I found the 50Mp back easier to use (back horizontal/with rise fall) than with the back horizontal (shift L/R). I tend to do more L and Right movements, with tilt in my work. The older Sony showed the faint banding more when the back was horizontal/landscape and shifted left and right. I did not find it a problem enough not to keep the back, but other reasons, kept me at the 260.
Interesting on the highlights issues, but that seems to be the case with CMOS, in general in that I tend to expose for highlights and realize I have more shadow room in post, just the opposite of how I work with the 260.
Thanks again for the images.
Just to be clear, I set up the camera with 18mm of back fall. Then I mounted the back in both vertical and horizontal positions so I could see the difference between the back being shifted both in its horizontal direction and the vertical direction.
So all the images in landscape orientation had the back horizontal and moved down 18mm (like you thought). All the images in portrait had the back vertical and moved down 18mm.
So you are seeing the two different movement directions. I did it this way on purpose so I could compare the same blue sky/white clouds.
What I haven't shown here is mounting the back in portrait orientation and shifting left/right. That does not stress the image circle as much as what I did here. I did some of that yesterday, but it was a dull blank cloudy sky so it wasn't much of a challenge for the LCC.
Here is what I mean about the highlights. I don't think this is any different that what others have shone before (Voidshatter's security camera). This is v9.0.3, and perhaps Phase will continue to make improvements.
Not that this is a deal killer, just something the 180 handles better. I agree it seems to be a CMOS thing. In general they seem more sensitive to purple. I'm sure there are 50 technical reasons why that is not the case, but it sure is my perception.
Edit: note that the 3100 image was even 1 stop under the 180 image. I matched exposure and color on the wall to the left. These are both 100% crops from the upper left of an image (40hr).
There is also another side note: the more shadow push applied, the more purple fringing appears. If you bracketed for your test shots, you could also try a brighter shot of the IQ3100 image - it may have less CA.
It is strange though that neither the IQ380 nor the IQ250 had such purple glow.
In the example is the 180 image on the right of the two vases? I am assuming this so due to shot count?
Both images show it the image in the left obviously has it in a greater amount and it's brighter.
Wonder if the defringe tool in LR would remove this effect. C1 has no such eye dropper tool that I am aware of.
davechewphotography.com1 Member(s) liked this post
Again this isn't a big deal. As Paul says I think defringe will take care of it. It is a curious difference though.
Is it possible to get a test of shift+rise combined to the edge/beyond the image circle on a wide angle, comparing the x80 and the 100mp backs? I have yet to see a well controlled test anywhere that is looking at the stitching potential. 60XL would be ideal (or 43xl)
1 Member(s) thanked for this post
This may be a separate issue from what "Void" has seen. It may just be a problem of highlight recovery. Would I have the raw files I would look at the specular highlights and how each channels goes into/out of clipping.
It would be possible that the issue may be some kind of chromatic aberration. Lloyd Chambers has seen similar stuff on the Leica S, but he found out by accident that they were not visible when exposure was by white led. After that he found out that the effect could be eliminated by an IR cut of filter. (Or was it UV cut off?)
It may also be that it is caused by some "feature" of the CMOS sensor. It is not the same design that is used in IQ-250.
If you would be kind enough to share the raw images it would be most helpful.
Here is the link to the raw files:
When I originally made the switch from a Canon 5D to the IQ180, I felt the way the 180 handled this type of highlight was more natural. As you say, it could be several things. The exif data says it was a 2 sec exposure on the 180 and a 1 sec on the 3100. If anything the 180 image should be clipped more given the longer exposure and the lower DR. But as you know the recorded shutter speed on these backs is not all that accurate when using a copal shutter, so I can't say for sure the 180 image was longer. They were both set to ISO 100 and the same f-stop.
The lighting was difficult on purpose, but identical for both. In fact the lens flare was pretty close to identical in each shot. I used to refer to this purple fringing from saturated areas as "blooming", but I'm not sure that is the right term. I looked at a similar shot from the a7rII on a Batis 25. It handles this better than the 3100 but not as well as the 180. Again many different variables so who knows.
I will repeat this is more of a curiosity than any sort of "issue"; it just isn't that different to worry about. For me, the decision to upgrade will basically be 1) live view and 2) staying on the upgrade discount path vs. the cost of doing so. Very rarely do I want to shift/rise/stitch anything wider than my 60xl, and the 3100 seems to handle that just fine up to 18mm. And unlike Voidshatter I have no interest in very long exposures or in pushing shadows beyond minor adjustments. So (again for me) any upgrade will have nothing to do with image quality.
davechewphotography.com1 Member(s) thanked for this post1 Member(s) liked this post
Thanks for the links.
The highlight area, is actually very small at 100% view, and may not even be noticed, but it does easily clean up in LR, with the Definge tool. C1 can't really get it with the CA adjustment or the manual defringe slider.
The 380 does a good job on the subject, however the areas that immediately standout are the foreground lower corners, whee the 2 seats are, as you have no issues pulling up the shadows and keeping details on the CMOS shot. It's interesting to play around inside these files, to realize the great depth that both backs can give. Overall the 380 file looks great, until you open the 100MP file and look into spots of shadow, where the 380 starts to give up details. But to be fair to the 380, with C1 9.03 IMO the files are very close, assuming both were at base iso?
I see better details on the stereo, and the set of folders in the lower right corner and a bit more details on the horse painting.
Thanks for sharing.
Thanks for the files!
I think it is a clipping issue. If you check the enclosed screen dumps you can see that IQ3-100MP has fewer clipped pixels in the red channel then in the other channels. The IQ-180 is almost perfectly balanced.
I guess this is caused by the IQ3-100MP having a different balance between sensivity on the blue and red channels compared to the IQ-180. The reason may be that the IQ3-100MP is more balanced for good performance in mixed (low colour temperature) light while the IQ-180 may be more balanced for daylight/strobe. Call it a guess or possibly an educated guess.
Thanks for doing that Erik. It is interesting that the red channel looks almost identical in both. You say the red Chanel is less clipped, which I understand. But the y axis is scaled differently, so I think the red channels are very close between the backs.
However, the green and blue channels are almost a whole stop brighter in the 3100 file. Note the natural light from windows off to the right is a lot brighter in the 3100 file. Could be just clouds or a different sensitivity. The 180 file was the longer exposure.