Ad
Ad
Ad
Pages: « 1 ... 4 5 [6]   Bottom of Page
Print
Author Topic: Striations in P45+@800 ISO normal ?  (Read 20049 times)
eronald
Sr. Member
****
Offline Offline

Posts: 3645



WWW
« Reply #100 on: November 28, 2007, 03:55:32 PM »
ReplyReply

Ok, so we guys with Kodak chips had better find a workaround.
Could you please yousendit.com the Raws ?

Edmund

Quote
Yes, and nice greenish spots...   

To me the P45+ came out as having a slightly lower dynamic range (very, very, very slightly.) But I guess Panopeeper should get his +4 stop shots soon and we can all get the numbers out all solid and shiny! (anyone up for this, I was supposed to do them, but forgot!) And now the nice H3D-39 is on its way back...

-axel
[a href=\"index.php?act=findpost&pid=156755\"][{POST_SNAPBACK}][/a]
« Last Edit: November 28, 2007, 03:56:26 PM by eronald » Logged

Edmund Ronald, Ph.D. 
godtfred
Sr. Member
****
Offline Offline

Posts: 293



WWW
« Reply #101 on: November 28, 2007, 06:05:27 PM »
ReplyReply

Quote
Could you please yousendit.com the Raws ?
[{POST_SNAPBACK}][/a]
Here are the raws:

[a href=\"http://download.yousendit.com/0D3CED025512EFCC]http://download.yousendit.com/0D3CED025512EFCC[/url]

-axel
Logged

Axel Bauer
godtfred.com
H2|M679CS|P45+
eronald
Sr. Member
****
Offline Offline

Posts: 3645



WWW
« Reply #102 on: November 28, 2007, 06:10:06 PM »
ReplyReply

The stripes move around with different Raw software, so a workaround should be possible - I've sent Brian of Raw Developer an email, maybe registered users or otthers would find it useful to send him a note or a bounty note too. I think he might be in a good position to solve this or at least tell us exactly what's going on.

Edmund

Quote
Here are the raws:

http://download.yousendit.com/0D3CED025512EFCC

-axel
[a href=\"index.php?act=findpost&pid=156782\"][{POST_SNAPBACK}][/a]
Logged

Edmund Ronald, Ph.D. 
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #103 on: November 28, 2007, 10:39:54 PM »
ReplyReply

Quote
The hassy file has to go through .dng conversion to be operable in ACR, this loses a lot of the hassy data like color accuracy etc. (or so hassy claims... and is correct in my experience with different hassy systems over the last couple of years.)

Whose DNG converter is that? Is it their own production? Where is that to get from?

I can do nothing with the .FFF file. I don't know the format (it is in a TIFF container, but that's not enough). Adobe's DNG converter can't process it (of course; if it could, then ACR would accept it as well).

Anyway, it sounds to be unbelievable, that the DNG conversion loses image data. Color accuracy is no separate information in the raw image, it can not be "lost". There may be another problem: the in-factory calibration of ACR is less than optimal for several cameras. It can be calibrated to yield matching color, or anything you want to. Those, who are very picky about colors, calibrate their camera themselves, because there can be small variations between copies of the same model.

Another problem may be, that their DNG converter generates incorrect color conversion values (in the DNG file).

Quote
I guess Panopeeper should get his +4 stop shots soon and we can all get the numbers out all solid and shiny!

I took another look at your previously posted raw images and found green pixels with the value 65536 (on the specular top edge of the Toblerone). Not many, only 3400 of them; that is is 0.0022%, too few to be shown by the histogram.

Still, this proves, that the sensor in fact delivers 65536 levels in ISO 400. Consequently the top level in the newer image, 001136, namely 15000 is still two stops from clipping:

Histogram of 001136

As the counters show, there are only about 2200 pixel over 15000, that's negligable.

Now, what does this mean in the praxis?

1. 18.7% of all pixels in 001136 are in the range 0-127. If we start at the right end (i.e. with the range of levels from 32768 to 65535) as +0EV, then the range 0-127 corresponds to -9 EV.

2. This shot is not so bad, but Edmund's CF000965 had 10.5% of all pixels with value zero (his other posted shot had 64%).

Now, Edmund may say that this is not a problem for him, he does not want to see details in the deep shadows, only graceful noise. However, this is not so simple. The proportion of the values of the three "kind" of pixels makes the color (with de-mosaicing, of course); this is obvious. What is perhaps not so obvious is, that when one or two colors are on level zero, then the proportion is gone to the dog: on one place it may be (0, 50, 80), on an even darker spot with half the lightness (0, 25, 40). In other words, different shades of the same color can result in vastly different colors. This can appear as color noise, which may be acceptable, typically on very fine structured surface like fabric/cloth, but it may manifest as off-color spots as well.
Logged

Gabor
godtfred
Sr. Member
****
Offline Offline

Posts: 293



WWW
« Reply #104 on: November 29, 2007, 01:07:14 AM »
ReplyReply

Quote
Whose DNG converter is that? Is it their own production? Where is that to get from?
[a href=\"index.php?act=findpost&pid=156845\"][{POST_SNAPBACK}][/a]
It is a function of Flexcolor to export DNG's. Hasselblads info on this (anyone correct me if im wrong) is that there are headers and exif fields in the .3FR files (the backbone, before they become .fff) containing vital info for the use of DACII, its file correction software. Also I believe Flexcolor downloads some sort of "black-subtraction frame" or similar from the back of each camera being attatched to it. This does not follow through to other apps, consequently not applying all possible file corrections.

Personally I don't know if any of this is correct, but I have through my some ten thousand files with the H3D tried .dngs several times to recover highlights, and the images have had more noise and other problems not seen in Flex.

-axel
Logged

Axel Bauer
godtfred.com
H2|M679CS|P45+
TechTalk
Sr. Member
****
Offline Offline

Posts: 295


« Reply #105 on: November 29, 2007, 01:23:53 AM »
ReplyReply

Quote
The H3D-39 vs. P45+ ISO400 test, as promised!

And also as promised I'll eat my words           

Both images using the same lens (100mm f2.2) P45+ on a H2 body, focus distance is the same, and the cameras where tripod mounted with mirror up. Same exposure for both (f5.6, 1/80).

Everything is zeroed in every software (C1, Flex, ACR.) Every software is latest non-beta version. Each image has its own histogram pasted on top... The same levels type adjustment to bring out any problems is used.

I'll make raw files available in a couple of days to those who would like them... (please say so on the board, if there are fewer than 1 taker, I won't go to the trouble.)

enjoy. -axel
[attachment=4066:attachment]

EDIT: The hassy file has to go through .dng conversion to be operable in ACR, this loses a lot of the hassy data like color accuracy etc. (or so hassy claims... and is correct in my experience with different hassy systems over the last couple of years.)
[a href=\"index.php?act=findpost&pid=156740\"][{POST_SNAPBACK}][/a]
First question... Why did you choose a non-standard set-up in FlexColor when you shot and also saved the image? In both cases the "Film Response" curve is set to "Low Contrast" instead of "Standard or "Linear". Seems an odd choice. I've never used it for any image.

Second question... What is the point of this exercise of crushing and destroying image data and quality with radical compression and gamma settings? It does show that you can create image problems that would not exist otherwise, but other than that what are we learning?

I've attached a png file (from the H3D RAW) that has zero noise bias or color noise setting. It simply has a standard film response setting and the black point set to the most conservative end point.

[attachment=4075:attachment]
« Last Edit: November 29, 2007, 01:30:12 AM by TechTalk » Logged
eronald
Sr. Member
****
Offline Offline

Posts: 3645



WWW
« Reply #106 on: November 29, 2007, 01:33:07 AM »
ReplyReply

Edmund is going to say "how really astonishing that we have total zero values in an image file!"

That piece of cloth was imaged on an average Paris day, in room light where I could clearly see. There is texture to be seen on the cloth. No way the physics of the situation could push all those pixels on the cloth down to absolute zero.

Some sort of firmware cleanup seems to be going round and doing some sweeping up here, methinks.

Edmund


Quote
2. This shot is not so bad, but Edmund's CF000965 had 10.5% of all pixels with value zero (his other posted shot had 64%).

Now, Edmund may say that this is not a problem for him, he does not want to see details in the deep shadows, only graceful noise. However, this is not so simple. The proportion of the values of the three "kind" of pixels makes the color (with de-mosaicing, of course); this is obvious. What is perhaps not so obvious is, that when one or two colors are on level zero, then the proportion is gone to the dog: on one place it may be (0, 50, 80), on an even darker spot with half the lightness (0, 25, 40). In other words, different shades of the same color can result in vastly different colors. This can appear as color noise, which may be acceptable, typically on very fine structured surface like fabric/cloth, but it may manifest as off-color spots as well.
[a href=\"index.php?act=findpost&pid=156845\"][{POST_SNAPBACK}][/a]
Logged

Edmund Ronald, Ph.D. 
josayeruk
Guest
« Reply #107 on: November 29, 2007, 01:34:26 AM »
ReplyReply

Good point Tech Talk!  

Also if you look at the crest of the tape measure the blacks are much smoother on the H3D, in FlexColor and also in ACR but not as pleasing.

Still looks cleaner than the P45 though.

Jo S.x
« Last Edit: November 29, 2007, 01:36:36 AM by josayeruk » Logged
TechTalk
Sr. Member
****
Offline Offline

Posts: 295


« Reply #108 on: November 29, 2007, 01:35:59 AM »
ReplyReply

A couple of 100% crops from the previous image. The 100mm "H" lens is a stunning performer. I've always enjoyed using it. I like the reflections of the tape measure rule lines in the base of the orange.

[attachment=4076:attachment]  [attachment=4077:attachment]
« Last Edit: November 29, 2007, 01:40:21 AM by TechTalk » Logged
godtfred
Sr. Member
****
Offline Offline

Posts: 293



WWW
« Reply #109 on: November 29, 2007, 01:58:06 AM »
ReplyReply

Quote
First question... Why did you choose a non-standard set-up in FlexColor when you shot and also saved the image? In both cases the "Film Response" curve is set to "Low Contrast" instead of "Standard or "Linear". Seems an odd choice. I've never used it for any image.
[a href=\"index.php?act=findpost&pid=156877\"][{POST_SNAPBACK}][/a]
Sorry, my bad! I have inadvertently used my "low contrast profile" in flex instead of setting from the default (where i knew id get something else than ProfotoRGB as the profile. Linear should be the right choice (I'll refresh my old post during the week with this setting.)

Quote
Second question... What is the point of this exercise of crushing and destroying image data and quality with radical compression and gamma settings? It does show that you can create image problems that would not exist otherwise, but other than that what are we learning?
[a href=\"index.php?act=findpost&pid=156877\"][{POST_SNAPBACK}][/a]
The point of this exercise for me, was to verify at what limit I would have to retouch heavily if i underexposed using high ISO shots (low iso has never been a problem.) ISO800 on the P45+ is a lot worse than 400, yet as the H3D does not go to 800 yet, the test was at 400 to look for ANY striations and compare for reference to a faulty P45+, not to give any camerasystem any type of poor performance rating (they are both excellent.)

The test has for me shown that I can not (in any similar way) utilize the same high dynamic range to recover faulty (underexposed) images as I have done on some occasions with ISO50-100 shots, whithout a penalty of some magenta casts/lines. These do not appear at lower iso, and i can "pull out the shadows" to a much greater extent.

-axel
Logged

Axel Bauer
godtfred.com
H2|M679CS|P45+
TechTalk
Sr. Member
****
Offline Offline

Posts: 295


« Reply #110 on: November 29, 2007, 02:12:54 AM »
ReplyReply

Quote
The test has for me shown that I can not (in any similar way) utilize the same high dynamic range to recover faulty (underexposed) images as I have done on some occasions with ISO50-100 shots, whithout a penalty of some magenta casts/lines. These do not appear at lower iso, and i can "pull out the shadows" to a much greater extent.

-axel
[a href=\"index.php?act=findpost&pid=156886\"][{POST_SNAPBACK}][/a]
I don't think that taking a properly exposed image and then throwing out a huge amount of levels in the mid-tone to shadow range is a good way to simulate what can be recovered in an underexposed image. You might want to duplicate your experiment by actually under exposing at whatever ISO you want to test.
Logged
godtfred
Sr. Member
****
Offline Offline

Posts: 293



WWW
« Reply #111 on: November 29, 2007, 02:41:50 AM »
ReplyReply

Quote
A couple of 100% crops from the previous image. The 100mm "H" lens is a stunning performer. I've always enjoyed using it. I like the reflections of the tape measure rule lines in the base of the orange.
[a href=\"index.php?act=findpost&pid=156881\"][{POST_SNAPBACK}][/a]
I completely agreen on the performance of this lens, I had around 10000 exposures on it when I checked a couple of months back, and it has not failed me in any situation yet. I find all my lenses top notch (35, 100, 120 and 210) but don't like the 35 a lot, maybe because I have never been very good at wide angle shots...    

Quote
I don't think that taking a properly exposed image and then throwing out a huge amount of levels in the mid-tone to shadow range is a good way to simulate what can be recovered in an underexposed image. You might want to duplicate your experiment by actually under exposing at whatever ISO you want to test.
[a href=\"index.php?act=findpost&pid=156888\"][{POST_SNAPBACK}][/a]
I don't either, but if you follow my other posts, you will see that I have posted underexposed raw files from the P45+. The last test between the H3D-39 and P45+ was a continuation of this mainly for my own personal interest and to see if the P45+ had a fault. (I had never experienced any striations like this on my H3D, so had to see for myself, guess I just don't underexpose much   )

Using linear (that I now found out i did not do!) was to make the comparison as equal as possible in the black's, as C1 has a very aggressive tone curve, and nothing in between. This would render subsequent comparison with levels in PS useless. (It might be useless anyway, but for myself I found out a couple of things along the way   )

A really, really important thing this has shown me, is how much C1 does before you even start editing, this is the opposite approach when compared to Flexcolor, and one that I don't fancy. I'd rather add than subtract in my image editing.

Quote
Also if you look at the crest of the tape measure the blacks are much smoother on the H3D, in FlexColor and also in ACR but not as pleasing.

Still looks cleaner than the P45 though.
[a href=\"index.php?act=findpost&pid=156880\"][{POST_SNAPBACK}][/a]
+1

-axel
Logged

Axel Bauer
godtfred.com
H2|M679CS|P45+
Ray
Sr. Member
****
Offline Offline

Posts: 8812


« Reply #112 on: November 29, 2007, 09:06:43 AM »
ReplyReply

It is a worry, isn't it!   . We all know that MFDBs are not too hot at high ISO. That means, if you want to get decent shadows, you've got to push the exposure to it's maximum and perhaps blow a few largish specral highlights which you otherwise wouldn't.

I don't see any rocket science here. There has to be at least some QC variation in DB quality from copy to copy. I exchanged my first copy of my 5D because I thought banding was unacceptable in deep shadows. But I have to admit, the image that first brought the problem to my attention was slightly underexposed.

The second copy was, I thought, marginally better regarding shadow banding but it seemed clear to me that the first camera was not really faulty. It was just an example of QC variation which one can sometimes come across in lenses. Nothing's perfect.
Logged
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #113 on: November 29, 2007, 11:19:15 AM »
ReplyReply

Quote
Edmund is going to say "how really astonishing that we have total zero values in an image file!"

Well, 10% is not astonishing (though the 64% is), but it is strongly underexposed in my books.

Quote
There is texture to be seen on the cloth. No way the physics of the situation could push all those pixels on the cloth down to absolute zero

Pleeeze Edmund, I mentioned 10.5%, not all those pixels. See the figures for the following selection (marked by red rectangle):

crop from CF000965

Btw, this one shows, that there is no pixel over 7600; this is for the entire image, not for the selection. 7600-15200-30400-60800, it is over three stops from the right edge, 65535.

If you expose it two stops higher and reduce it in post proc to the same darkness, it can "gracefully" sink into black without that much color noise.
Logged

Gabor
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #114 on: November 29, 2007, 11:39:03 AM »
ReplyReply

Quote
I completely agreen on the performance of this lens, I had around 10000 exposures on it when I checked a couple of months back, and it has not failed me in any situation yet

This is not the current subject, I would not have mentioned it, but I noticed something interesting on these shots. Following are 100% crops from three of your posted raw images. 001127 is the ISO 800 shot; all three are sharper, than the ISO 400 shots. The difference is observable on other parts as well.

What you see on these crops is not only unsharpened but not de-mosaiced, each pixel on its own (strongly green, for not not even white balanced). Only the sensor and the AA filter are between the lens and these crops.

Crops from 001123, 001124 and 001127
Logged

Gabor
godtfred
Sr. Member
****
Offline Offline

Posts: 293



WWW
« Reply #115 on: November 29, 2007, 01:39:54 PM »
ReplyReply

Quote
This is not the current subject, I would not have mentioned it, but I noticed something interesting on these shots. Following are 100% crops from three of your posted raw images. 001127 is the ISO 800 shot; all three are sharper, than the ISO 400 shots. The difference is observable on other parts as well.

What you see on these crops is not only unsharpened but not de-mosaiced, each pixel on its own (strongly green, for not not even white balanced). Only the sensor and the AA filter are between the lens and these crops.

Crops from 001123, 001124 and 001127
[a href=\"index.php?act=findpost&pid=156995\"][{POST_SNAPBACK}][/a]
Is this not a function of wider DOF? The one extra stop closes down the aperture more and has a large impact at such a close focusing distance?

-axel
« Last Edit: November 29, 2007, 01:40:28 PM by godtfred » Logged

Axel Bauer
godtfred.com
H2|M679CS|P45+
Panopeeper
Sr. Member
****
Offline Offline

Posts: 1805


« Reply #116 on: November 29, 2007, 04:42:47 PM »
ReplyReply

Quote
Is this not a function of wider DOF? The one extra stop closes down the aperture more and has a large impact at such a close focusing distance?

I forgot that 100mm on your camera means a much wider FoV than on the APS-C sensor of my camera. With a 100mm lens I would have to go much farther away (85mm is already a tele lens for me).
Logged

Gabor
Pages: « 1 ... 4 5 [6]   Top of Page
Print
Jump to:  

Ad
Ad
Ad