The GetDPI Photography Forum

Great to see you here. Join our insightful photographic forum today and start tapping into a huge wealth of photographic knowledge. Completing our simple registration process will allow you to gain access to exclusive content, add your own topics and posts, share your work and connect with other members through your own private inbox! And don’t forget to say hi!

IQ250 movement tests on the 23/32/40/70/90 - desaturation and mazing artifact

torger

Active member
Mazing occurs because the demosaicer expects green 1 and green 2 to be the same on a uniform surface, but due to crosstalk they separate and then the demosaicer starts to make up detail. The greens differs because one of the greens get most it leaks from the neighboring red and the other from the blue. Of course if you're lucky and the color you're shooting has equal red and blue content (on the raw level), green1&2 will not separate.

As Doug says it's not only about desaturation and mazing, crosstalk means that you have mixing of color channels and thus worse color separation. Degradation in color separation can be really hard to see, but if you're in medium format because of color-optimized CFA and other subtle advantages in terms of color you should think twice before wasting that advantage by mixing the channels with crosstalk.

If you want to it's really easy to take away the mazing, you just equalize the green channels. Lumariver HDR does this, and you can do it with the green equiliberation feature in RawTherapee. It's probably only a matter of time before Phase One implements it in Capture One. On the other hand mazing is a good "health indicator" meaning that you're pushing the sensor too far. If you're making a monochrome shot it will work out though.

I agree that Phase One, Hassy etc should make a thorough test of tech lenses inhouse and make the results publically available for all dealers and users to see. It's a better model than leaving it to a few select dealers and laymen. You may need to have inside information to make the tests correctly, for example with the Sony sensor it was unknown for a long time if it had offset microlenses or not (it has), which affects how the tests should be performed.
 

Paul2660

Well-known member
Mazing occurs because the demosaicer expects green 1 and green 2 to be the same on a uniform surface, but due to crosstalk they separate and then the demosaicer starts to make up detail. The greens differs because one of the greens get most it leaks from the neighboring red and the other from the blue. Of course if you're lucky and the color you're shooting has equal red and blue content (on the raw level), green1&2 will not separate.

As Doug says it's not only about desaturation and mazing, crosstalk means that you have mixing of color channels and thus worse color separation. Degradation in color separation can be really hard to see, but if you're in medium format because of color-optimized CFA and other subtle advantages in terms of color you should think twice before wasting that advantage by mixing the channels with crosstalk.

If you want to it's really easy to take away the mazing, you just equalize the green channels. Lumariver HDR does this, and you can do it with the green equiliberation feature in RawTherapee. It's probably only a matter of time before Phase One implements it in Capture One. On the other hand mazing is a good "health indicator" meaning that you're pushing the sensor too far. If you're making a monochrome shot it will work out though.

I agree that Phase One, Hassy etc should make a thorough test of tech lenses inhouse and make the results publically available for all dealers and users to see. It's a better model than leaving it to a few select dealers and laymen. You may need to have inside information to make the tests correctly, for example with the Sony sensor it was unknown for a long time if it had offset microlenses or not (it has), which affects how the tests should be performed.
Hopefully if a full frame CMOS comes in the future, Phase One can work out a new software algorithim to help with these issues. I would hope that they are working on it now. I would hope that they either purchase the rights to the development that has already been down by your company, or they make their own software fix. As it seems this full frame is a year or so away, maybe they can make changes to effect the current 50MP CMOS. Of course as Phase is a very closed company, it's hard to know if they even agree that mazing is an issue and even if it exists.

Paul
 

tjv

Active member
I will say that mazing is quite easy to see on the DT library test, so it's not only a big problem on flat subjects. With the test shots that Darr kindly shared, it's also visable, particularly after 10mm of horozontal shift.
I keep flip flopping between wanting to buy the 50c and CCD 50 Hasselblad, and these tests are brilliant in helping me understand what I could expect in terms of performance. Thanks!
 
I will say that mazing is quite easy to see on the DT library test, so it's not only a big problem on flat subjects. With the test shots that Darr kindly shared, it's also visable, particularly after 10mm of horozontal shift.
I keep flip flopping between wanting to buy the 50c and CCD 50 Hasselblad, and these tests are brilliant in helping me understand what I could expect in terms of performance. Thanks!
I have just quickly checked the Rodenstock RAW files provided by the DT test (i.e. the library interior shot) and within my suggested amount of movement I did not observe any mazing (if I did not miss it). Could you please point out where you observe mazing within the movement range I specified?

Also I downloaded the RAW files provided by Darr but I am still not sure where the mazing is (and I thought it was not Rodenstock but Schneider if I remember correctly?). Could you point out which part of which file you cropped?





I still stand by my hypothesis that for the Rodenstock wide angles the worst case for mazing is only the blue part in the LCC shot, and is only affected by movements along the shorter edge of the sensor. If you refrain movements beyond the range I suggested in that direction then there should be no mazing. I proposed this hypothesis by systematical testings and I am yet to see Doug providing a counterexample.
 
Last edited:
Here are the library test shots for the 40HR. Again, mazing only happens when movements along the shorter edge of the sensor exceeds my suggested range.

 
The mazing I saw is not multi coloured like what you have highlighted above, rather presents itself quite obvious (to me, at least) and strange maze like patterns.
I posted examples here: http://www.getdpi.com/forum/621422-post90.html
Perhaps I'm being too picky?
From the screenshots you posted:

a) The mazing on the 32HR comes from a stitched JPEG, and that area comes from an exposure with movement exceeding my suggested range, i.e. it's from the 30mm rise one. I have posted above showing you that if you use 15mm rise you will never get that mazing there. (Same crop position as yours: see #25 of this thread)

b) The 47XL picture was taken by a Schneider lens and I cannot comment on that.
 

tjv

Active member
This is below the 15mm shift line.
I know it's at 200% magnification, but for some reason I can't attach screen grabs and have them show bigger than a thumbnail... I had to downsample to 1200px wide...
Anyway, I find the strange maze like pattern in the flat(it) areas of tone disturbing. It's most evident in the mid brown wood tone, but also shows up in what looks like false / overly jagged edges around details.
Maybe someone could tell me what I'm looking at and if it's related to the more crazy mazing that can be seen in other areas of the stitch?
 
This is below the 15mm shift line.
I know it's at 200% magnification, but for some reason I can't attach screen grabs and have them show bigger than a thumbnail... I had to downsample to 1200px wide...
Anyway, I find the strange maze like pattern in the flat(it) areas of tone disturbing. It's most evident in the mid brown wood tone, but also shows up in what looks like false / overly jagged edges around details.
Maybe someone could tell me what I'm looking at and if it's related to the more crazy mazing that can be seen in other areas of the stitch?
You still did not get my point. You are looking into a JPEG file, and the JPEG file is a stitched file. The cropped part you are looking at 200% is actually from a RAW file taken when the lens was shifted 30mm upwards even though it's below the 15mm line. Doug did not have prior knowledge of this mazing artifact when he stitched the JPEG file.

You would need to download Doug's RAW files, and look into the RAW file taken when the lens was shifted 15mm upwards. It would still cover the area of your interest (as I posted in #25 of this thread) and you will not be able to see such mazing artifact there (on the same subject in that library, but in a different position in the exposure as the movement was different).

If Doug knew this mazing artifact well, then he would have chosen to stitch from the 15mm-movement images. He could still achieve the same stitch if he shoot with the back in portrait direction when he does fall of the back to shoot towards the ceiling. (This is just a hack / workaround for the mazing artifact because the sensor would only give mazing when movement along the shorter edge of the sensor exceeds a certain amount.)

Below shows the hack to avoid mazing and to achieve the same stitch: I had to borrow the IQ140 to illustrate the portrait movements as well, but take all these 4 overlapping rectangles as IQ250s. In such way the sensor movement would never break the rule that the movement along the shorter edge of the sensor should not exceed 15mm (actually, 12mm to be conservative for the 32HR as suggested in the OP). This would work to avoid mazing if my hypothesis is true.

 
Last edited:

tjv

Active member
Ah, I see. Thanks for clearing that up. I've tried multiple times to download the RAW files, but it seems the DT server (or the link to my country, more like) is so slow that I leave it downloading all night and don't get half way though one file!
Regardless, what you say makes sense and is very heartening. I'm hoping to get the go ahead to buy an H5D50c kit this week at work, so I guess I'll know for myself soon enough.
 

tjv

Active member
PS: Next time you're in contact with Paula at Linhof Studio, and if you're interested, it'd be good to shoot the same shots on the CFV-50c she has to demo and your IQ250. I'm really interested to see if Phocus and C1 are equal with regards to mazing.
 

torger

Active member
There's an additional type of artifact which I have just recently become aware of, vertical banding similar to microlens ripple but much smoother, more irregular and lower frequency (about 16 pixel period).

This banding can for example be seen on the sides in the DT library test on the 32mm HR where the sensor is shifted sideways (in landscape orientation). I have not yet tested if C1 cancels it out, but I would guess that it does.

I have a CFV-50c test file where I can see the same issue, and as Phocus LCC is simple and only makes a blur it will not cancel that out.

I previously thought Phocus was good with the CFV-50c as the Sony sensor don't show any tiling or ripple like the Dalsa 6um, but apparently I was wrong. This banding will however only be visible in ultra-high contrast processing such as sometimes used in black&white or other types of artistic processing. You will probably have less problem with this than with tiling on Dalsa sensors.

I've attached a 100% crop of a LCC shot with strongly increased contrast which shows the vertical banding.
 

tjv

Active member
Interesting, although heartening to know it only presents itself with extreme processing. I wonder if Phocus will get an upgrade in the scene calibration feature to better deal to these kinds of things?
With regards to that, how's the programming going for your own RAW converter? Are you making headway with regards to better support for the CMOS backs?
 

torger

Active member
A one-liner fix for them would be the reduce the blur radius to say 4 pixels or so (I would guess it's 50-100 today), dust spots would disappear and also that banding and radius is still large enough to kill most noise in the LCC shot.

Due to various other missions I'm not working that much on Lumariver HDR now, but I'm planning to make an upgrade of the raw conversion layer soon(ish) and then CFV-50c will be included. At some point I will probably look into making Phocus compatible raws too (I'm a Hassy user myself now, so I see a need ;) ), so when you come across something difficult you can first cook up a raw in Lumariver HDR with corrections and reconstructions and then make a raw conversion as usual in Phocus.
 

tjv

Active member
Anders, would that also help with reducing the tiling problem on some Delsa 60 shots?

Off topic – I see Hasselblad are now advertising quite significant reductions on the H5D-50 camera and back, stating that it's end of line clearance. Kind of a sad day hearing the most tech-cam friendly sensor will no longer be sold.

Back on topic – The more I read about the Sony CMOS sensor in any of it's implimentations the more I am confused as to whether it would be a good option for my work. I really, really wish I could borrow a back adaptor for my Techno and demo a CFV-50c. The latter is possible, the former near on impossible.
 

torger

Active member
Anders, would that also help with reducing the tiling problem on some Delsa 60 shots?

Off topic – I see Hasselblad are now advertising quite significant reductions on the H5D-50 camera and back, stating that it's end of line clearance. Kind of a sad day hearing the most tech-cam friendly sensor will no longer be sold.

Back on topic – The more I read about the Sony CMOS sensor in any of it's implimentations the more I am confused as to whether it would be a good option for my work. I really, really wish I could borrow a back adaptor for my Techno and demo a CFV-50c. The latter is possible, the former near on impossible.
It would not help the Dalsa, its tiling and its microlens ripple is sharp so you need a much smarter LCC algorithm to fix that. The CFV-50c ripple/banding is smooth so you can get away with a simple blur if the radius is not too big.

(Sad to hear on the H5D-50, but expected. I was surprised that it didn't happen earlier. People want CMOS now, in a H-system DSLR it comes to it's full right. However the clearance is good for people getting gear below the highest end, great deals can be had on the 50MP CCDs and with Schneider Digitar lenses you have a very high performing and flexible tech system. I still hope that future CMOS will fix the angular response issue so I can go from the 50MP CCD to a newer generation CMOS with wide anglar response in a few years. That scenario is not too unlikely.)

Is the adapter plate the only problem for the demo? Isn't it possible to get a V-mount adapter plate for your sliding back?
 
There's an additional type of artifact which I have just recently become aware of, vertical banding similar to microlens ripple but much smoother, more irregular and lower frequency (about 16 pixel period).

This banding can for example be seen on the sides in the DT library test on the 32mm HR where the sensor is shifted sideways (in landscape orientation). I have not yet tested if C1 cancels it out, but I would guess that it does.

I have a CFV-50c test file where I can see the same issue, and as Phocus LCC is simple and only makes a blur it will not cancel that out.

I previously thought Phocus was good with the CFV-50c as the Sony sensor don't show any tiling or ripple like the Dalsa 6um, but apparently I was wrong. This banding will however only be visible in ultra-high contrast processing such as sometimes used in black&white or other types of artistic processing. You will probably have less problem with this than with tiling on Dalsa sensors.

I've attached a 100% crop of a LCC shot with strongly increased contrast which shows the vertical banding.
Phase One doesn't cancel this artifact in Capture One v8.1 yet. I will post examples.
 
I can no longer edit the original post so I can only make an additional note here from Linhof Studio:

‘Linhof Studio were pleased to collaborate with these tests to give the photographers as unbiased opinion and help with informed decisions in their purchases. The Linhof Studio do not participate in Forums as they strongly believe this should be a free place for customers to discuss topics openly amongst themselves without the natural biased opinion of manufacturers or re-sellers. They do however offer excellent support and information on an individual basis if contacted. ‘
 

dougpeterson

Workshop Member
It would not help the Dalsa, its tiling and its microlens ripple is sharp so you need a much smarter LCC algorithm to fix that. The CFV-50c ripple/banding is smooth so you can get away with a simple blur if the radius is not too big.

(Sad to hear on the H5D-50, but expected. I was surprised that it didn't happen earlier. People want CMOS now, in a H-system DSLR it comes to it's full right. However the clearance is good for people getting gear below the highest end, great deals can be had on the 50MP CCDs and with Schneider Digitar lenses you have a very high performing and flexible tech system. I still hope that future CMOS will fix the angular response issue so I can go from the 50MP CCD to a newer generation CMOS with wide anglar response in a few years. That scenario is not too unlikely.)

Is the adapter plate the only problem for the demo? Isn't it possible to get a V-mount adapter plate for your sliding back?
Rippling on Dalsa sensors is partially or entirely removed when using the LCC tool in Capture One using the optional setting "Technical Camera Lens".
 
Top