Reading mode: Light Dark

G1 and GH2 pushed three stops.

Go to last post
dko22 said:
The G1 has reasonable DR at base ISO. But, as Louis says, try boosting the shadows at 200 on a GI and it's not pretty. I found that GH2 at ISO 320 was similar to G1 at 100 (and about 12 months ago posted one or two examples during the short time i had both cameras). In general it's fair to say the GH2 is about 1 1/2 stops better. Will the E-M5 equal or even surpass it? We'll soon find out...
Click to expand...
If you base your conclusion about 1,5 stops better on your comparison between the GH2 at ISO 320 versus the G1 at 100, you need to keep in mind that the GH2 calibrates its ISO scale differently than the G1. The G1 underestimates its ISO relative to the DxO definition whereas the GH2 does not. If you put the G1 on the same ISO scale as the GH2, the G1 begins at ISO 160 rather than 100, which reduces the difference between the two points you compared to one EV. The fact that the loss of DR is typically not directly proportional to the ISO gain at the beginning of the ISO scale (due to reduced read noise when the signal is amplified), may reduce the difference you inferred to slightly less than one EV.
 
Louis_Dobson said:
While I agree, I did it that way because they were both Pannys.
Click to expand...
Panasonic has considerable inter-model variability in terms of the relationship between nominal ISO and DxOmark-measured ISO. When left to autoexposure, the GH2 is more tuned to preserve highlights (measured ISO=nominal ISO), and the G1 is more tuned for shadow latitude (measured ISO> nominal ISO). If anything, that would bias your shadow test towards favoring the G1.
Louis_Dobson said:
In fact, all this tells me is what I already knew, by the most effective method possible - I have used both cameras and I already know that the G1 files need careful massaging, whereas the GH2 has enough margin for me to do whatever I choose.
Click to expand...
I agree that is what's most important, but what you've posted here is helpful for people who want to know the difference without discovering it for themselves. I've shot enough with these cameras to know that the data you presented are representative/legit.
 
I chose an arbitrary exposure for each to match highlight recoverability in order to have a controlled assessment of the shadows. That the GH2 ended up with slightly more recoverable highlights is an error in my testing method rather than an advantage of the GH2. I don't really believe that any camera has more recoverably highlights, intrinsically. One camera may have more DR than another, and we manually determine whether that DR is used for the highlights or the shadows. If we let the camera choose the exposure, then it will bias towards one or the other based on the manufacturer's intent.
Adventsam said:
GH2 appears to have more recoverable highlights than the K-5? in LR4, have a go yourself? its just a quick play so maybe not so, but my initial check was that it recovered all of the tarmac now better than the K5?
Click to expand...
 
Exactly. However, if the same file is irrecoverably clipped at both highlights and shadows, then we can get a sense for what Lightroom is doing by massive (eg 4 stop) pulls and pushes.
richarddd said:
We don't know what LR is doing, which seems to mean we have two unknowns, the sensor and the LR processing.

Image quality and need for manual processing could be due to sensor differences or differences in what LR is doing. I believe I'm just restating your post, but I want to make sure.
Amin Sabet said:
It's important to present the highlight latitude from the same files. There is no way to know what kind of tone curve Lightroom is applying by default without looking at both the shadows and the highlights. For example, if the G1 RAW file has more recoverable highlights in this comparison, then the shadows comparison is not apples-to-apples.

Classic example is the E-P1 vs GF1. E-P1 RAW files are underexposed by a stop relative to the GF1 RAW files (assuming auto exposure), and then the E-P1 files are "pushed" by a stop (relative to GF1 files) during JPEG processing in order to give both files the same final brightness. Lightroom gives the E-P1 and GF1 files the same treatment as their respective in-camera JPEG processing, ie Lightroom knows that the E-P1 files are underexposed (relatively) and applies a tone curve to "push" them.

As a result, the type of test you've done here (choosing f-stop based on nominal ISO) would show noisier shadows for the E-P1 and conclude lower DR for that camera than the GF1, whereas in actuality they have the same sensor.

I don't doubt that what you've shown here is real and true. However, DR testing is a pain in the neck and fraught with pitfalls.
Click to expand...
--
http://fruminousbandersnatch.blogspot.com/
Click to expand...
 
Amin Sabet said:
Louis_Dobson said:
While I agree, I did it that way because they were both Pannys.
Click to expand...
Panasonic has considerable inter-model variability in terms of the relationship between nominal ISO and DxOmark-measured ISO. When left to autoexposure, the GH2 is more tuned to preserve highlights (measured ISO=nominal ISO), and the G1 is more tuned for shadow latitude (measured ISO> nominal ISO). If anything, that would bias your shadow test towards favoring the G1.
Click to expand...
You are right that there is variability. But it's pretty easy to keep track of it. All the newer cameras (GH2, G3, GF3, GX1) are tuned "the GH2 way" (ISO starting at 160) and all the older ones are tuned the "G1 way" (ISO starting at 100).
 
Detail Man said:
Sagelight resizes with 16-bit arithmetic (but has been long stuck on a mandatory and rather at this point byzantine Bilinear re-sampling process, only). :P
Click to expand...
Ha.. Well, I have to agree with that. Sagelight's next version will have Lanzcos 3, 5, and 8 options.

Actually, now that I think about it, you're the one that talked me into it! Well, you were right -- it is much nicer, and I am pleased with the results.

Sagelight will be calculating the resizing at 64-bits per-channel. That seems to make a small difference.

Thanks again for convincing me on it...

Any thoughts on C*I*E LAB vs. RGB resizing?

Rob
 
rob nelson 2 said:
rob nelson 2 said:
Sagelight resizes with 16-bit arithmetic (but has been long stuck on a mandatory and rather at this point byzantine Bilinear re-sampling process, only). :P
Click to expand...
Ha.. Well, I have to agree with that. Sagelight's next version will have Lanzcos 3, 5, and 8 options.
Click to expand...
That sounds good, Rob ...
rob nelson 2 said:
Actually, now that I think about it, you're the one that talked me into it! Well, you were right -- it is much nicer, and I am pleased with the results.

Sagelight will be calculating the resizing at 64-bits per-channel. That seems to make a small difference.

Thanks again for convincing me on it...
Click to expand...
Thanks for attempting to address it.

BTW - I have always wondered if Sagelight's 16-bit Unsharp Masking tool performs luminance-only (or RGB) USM ? PaintShop Pro X4 (16-bit arithmetic pretty much in all functions) has the option to use either method (which is an option that I find is nice to have available, depending on image).
rob nelson 2 said:
Any thoughts on C*I*E LAB vs. RGB resizing?
Click to expand...
I am by no means well-schooled in this stuff. However, I have been doing some reading, and it appears that it is quite important to linearize any non-linear Gamma curves that are imposed on the image-data prior to conducting the re-sampling process , then re-apply relevant Gamma corrections.

This page has a lot of good information about that - and lists software (updated Jan 26, 2012) which does truly implement this evidently important technique. It appears that the 16-bit arithmetic applications Photoshop (CS4+), Lightroom (Versions 1.41+), PaintShop Pro (X2+), RAW Therapee, and Bibble (5+) do correctly perform such re-sampling using linearized image-data. See:

http://www.4p8.com/eric.brasseur/gamma.html

Note that (if linearizing sRGB), the Gamma constant is 2.4 (not 2.2). Thus, it would make sense (in the case of sRGB) to linearize the image-data using the specific sRGB encoding/decoding identities (which are given in the above-linked paper). Here is Color.org's full sRGB specification, as well:

http://www.color.org/srgb.pdf

From all appearances, CIE Lab is a gamma-encoded version of CIE XYZ that uses a Gamma constant of 3.0 . See these links for the relevant (CIE XYZ - CIE Lab) forward and reverse transformations:

http://www.getreuer.info/home/colorspace#TOC-CIE-Standard-Color-Spaces

and latter entries in: http://software.intel.com/sites/products/documentation/hpc/ipp/ippi/ippi_ch6/ch6_color_models.html

A couple of other pages which look to be informative (and accessible to the non-expert reader):

http://www.babelcolor.com/download/A%20review%20of%20RGB%20color%20spaces.pdf

http://www.cs.joensuu.fi/~pkoirala/article/RGB-space.pdf

Good luck with coding/development. Let us know when you finally do accomplish these things ...

DM ... :P
 
Might explain why the difference was less than I find in real life.

As I have said, this kind of "objective" testing is a poor second to actual experience. There are simply too many variables.
Amin Sabet said:
Louis_Dobson said:
While I agree, I did it that way because they were both Pannys.
Click to expand...
Panasonic has considerable inter-model variability in terms of the relationship between nominal ISO and DxOmark-measured ISO. When left to autoexposure, the GH2 is more tuned to preserve highlights (measured ISO=nominal ISO), and the G1 is more tuned for shadow latitude (measured ISO> nominal ISO). If anything, that would bias your shadow test towards favoring the G1.
Amin Sabet said:
In fact, all this tells me is what I already knew, by the most effective method possible - I have used both cameras and I already know that the G1 files need careful massaging, whereas the GH2 has enough margin for me to do whatever I choose.
Click to expand...
I agree that is what's most important, but what you've posted here is helpful for people who want to know the difference without discovering it for themselves. I've shot enough with these cameras to know that the data you presented are representative/legit.
Click to expand...
--
http://www.flickr.com/photos/acam
http://thegentlemansnapper.blogspot.com
 
I know what your saying and I'm on the same page, but lr4 has done something to m43 raw files, in my experience with the gh2, that it isn't necessarily doing with other raw files, with my exerience with the nex5n. Such that there was the ability to do things in lr4 with the 5n files that couldnt be done in lr3.6 but the recoverable highs and shadows didnt seem to change that much, with the GH2 its night and day better, so much so that its highlight pulling power seems exceptional.

But do try it yourself, so in many respects GH2 in lr4 just gained x stops in highlights, so if we had whatever dr previously how do we explain that it now has another 2 or 3 stops in highlights?
Amin Sabet said:
I chose an arbitrary exposure for each to match highlight recoverability in order to have a controlled assessment of the shadows. That the GH2 ended up with slightly more recoverable highlights is an error in my testing method rather than an advantage of the GH2. I don't really believe that any camera has more recoverably highlights, intrinsically. One camera may have more DR than another, and we manually determine whether that DR is used for the highlights or the shadows. If we let the camera choose the exposure, then it will bias towards one or the other based on the manufacturer's intent.
Adventsam said:
GH2 appears to have more recoverable highlights than the K-5? in LR4, have a go yourself? its just a quick play so maybe not so, but my initial check was that it recovered all of the tarmac now better than the K5?
Click to expand...
Click to expand...
 
Amin Sabet said:
Louis_Dobson said:
While I agree, I did it that way because they were both Pannys.
Click to expand...
Panasonic has considerable inter-model variability in terms of the relationship between nominal ISO and DxOmark-measured ISO. When left to autoexposure, the GH2 is more tuned to preserve highlights (measured ISO=nominal ISO), and the G1 is more tuned for shadow latitude (measured ISO> nominal ISO). If anything, that would bias your shadow test towards favoring the G1.
Click to expand...
The "measured ISO" on DxO is iso129 for the G1 and iso167 for the GH2, meaning that the G1 can handle app. 1/3 stop higher exposure at base ISO without clipping, and the G1 got 1/3 stop higher exposure (f/7.1 vs. f/8) in Louis' test, so everything should be fine.
Amin Sabet said:
Amin Sabet said:
In fact, all this tells me is what I already knew, by the most effective method possible - I have used both cameras and I already know that the G1 files need careful massaging, whereas the GH2 has enough margin for me to do whatever I choose.
Click to expand...
I agree that is what's most important, but what you've posted here is helpful for people who want to know the difference without discovering it for themselves. I've shot enough with these cameras to know that the data you presented are representative/legit.
Click to expand...
 
Earlier in this thread I wondered if better software (e.g., LR4) could make a sensor that seemed worse than another (e.g., when processed with LR3.6) now seem better. From your post, this seems quite possible or, at least, the gap between sensors could be narrowed substantially.

I wonder how LR4 does with the G3/GX1 sensor

LR4 is great software
Adventsam said:
I know what your saying and I'm on the same page, but lr4 has done something to m43 raw files, in my experience with the gh2, that it isn't necessarily doing with other raw files, with my exerience with the nex5n. Such that there was the ability to do things in lr4 with the 5n files that couldnt be done in lr3.6 but the recoverable highs and shadows didnt seem to change that much, with the GH2 its night and day better, so much so that its highlight pulling power seems exceptional.

But do try it yourself, so in many respects GH2 in lr4 just gained x stops in highlights, so if we had whatever dr previously how do we explain that it now has another 2 or 3 stops in highlights?
Click to expand...
 
richarddd said:
Earlier in this thread I wondered if better software (e.g., LR4) could make a sensor that seemed worse than another (e.g., when processed with LR3.6) now seem better. From your post, this seems quite possible or, at least, the gap between sensors could be narrowed substantially.

I wonder how LR4 does with the G3/GX1 sensor

LR4 is great software
Click to expand...
I think it will be similar, the problem with 3.6 is that as soon as you pull sky back for example it posterizes or color-shifts to purple and things. lr4 seems to bring the color back with no color shift or predict the colour. There is a good example on

http://www.photographyblog.com/reviews/panasonic_lumix_dmc_gx1_review/sample_images/

near the bottom there is couple of raw files with sky, one completely blown, I'll have a go myself later, but donwload them and try in 3.6 and 4, see if you see the difference in capability?
richarddd said:
Adventsam said:
I know what your saying and I'm on the same page, but lr4 has done something to m43 raw files, in my experience with the gh2, that it isn't necessarily doing with other raw files, with my exerience with the nex5n. Such that there was the ability to do things in lr4 with the 5n files that couldnt be done in lr3.6 but the recoverable highs and shadows didnt seem to change that much, with the GH2 its night and day better, so much so that its highlight pulling power seems exceptional.

But do try it yourself, so in many respects GH2 in lr4 just gained x stops in highlights, so if we had whatever dr previously how do we explain that it now has another 2 or 3 stops in highlights?
Click to expand...
Click to expand...
 
PatrickC said:
Big Ga said:
PatrickC said:
Big Ga said:
Extrapolating from these, I am tempted to say that the GH1 might indeed be the cleanest of the lot as DXO suggests - I think it could be with that white wall for example. However as other people have mentioned, the GH1 Achilles heel is banding. You might not see it in the white wall, but if there were large dark patches, it would probably show up.
Click to expand...
Banding in the GH1 can show up at ISO 1600, in underexposed areas. However I'm pretty sure "pushing" the exposure from a lower ISO, like was done here, would not show any banding, the pattern in the noise would not be there to start with.
Click to expand...
And I'm even more sure the banding is there from the word go.
Click to expand...
Do you even have a GH1? I do,
Click to expand...
"Do I 'even' have a GH1" eh ??
'even'...

Nice.
PatrickC said:
and it's pretty clear that pushing a low ISO image does not reveal any banding that was previously invisible.
Click to expand...
Is it really?

So ask yourself this. What does raising the ISO actually do? You yourself have said that at ISO1600 banding can show up. So where does this banding magically come from if it wasn't there in the first place?
 
Louis_Dobson said:
I think the downsizing is essential, a pixel to pixel comparison is meaningless....
Click to expand...
Hmmm. I'm not so sure. Sometime yes. Sometimes no.

Look at it this way - you might buy a 16MP camera to replace your 12MP one, because this lets you print slightly larger prints. However if you're interested in the quality 'per square inch' for example, then you don't re sample anything. You look at both at 100%
Louis_Dobson said:
Quite often a 2000th is my problem, differential development is my issue.

However, I will simulate a wedding for you later (I'll brush my hair and put on a tie).

what would you like, 1/15 wide open, pushed three stops?
Click to expand...
Now now. Remember that I have bipolar disorder when it comes to photography!!

I'm either interested in mega high ISOs, with mega thin DOF and super fast focus etc..
OR

I'm interested in the lowest of the low ISO. Extended DOF, and even the snail like focus of a 4/3 lens on a mft camera will mostly suffice!

Weddings and MFT aren't currently words (words??) that belong in the same sentence ;-)
 
Steen Bay said:
Steen Bay said:
I think you might be interested in the G3 ones I'll produce. I suspect you won't be happy, as they will probably show the G3 has visibly worse noise (at 100%) than the G1 at base ISO. And unlike others, I don't have faith that Oly will be able to 'tweak' the RAW files in any meaningful way (in fact, I'd be very, VERY worried that they'll do what they usually do, and sprinkle in a generous helping of nasty banding which will be in wait for those of us wanting to push and pull on the files)
Click to expand...
According to DxO the G3 beats the G1 in both (engineering) DR and SNR (18%), so it should have more 'usable DR' than the G1.
Click to expand...
I stand by what I said. And DxO agrees with me.
 
Big Ga said:
Steen Bay said:
Big Ga said:
I think you might be interested in the G3 ones I'll produce. I suspect you won't be happy, as they will probably show the G3 has visibly worse noise (at 100%) than the G1 at base ISO. And unlike others, I don't have faith that Oly will be able to 'tweak' the RAW files in any meaningful way (in fact, I'd be very, VERY worried that they'll do what they usually do, and sprinkle in a generous helping of nasty banding which will be in wait for those of us wanting to push and pull on the files)
Click to expand...
According to DxO the G3 beats the G1 in both (engineering) DR and SNR (18%), so it should have more 'usable DR' than the G1.
Click to expand...
I stand by what I said. And DxO agrees with me.
Click to expand...
According to DxO, the pixel-level DR is for practical purposes the same (10.06 vs 10.04) at base ISO. When the images are normalized to the same (8 MP) resolution, the G3 is slightly ahead (10.6 vs 10.3).
 
Anders W said:
Big Ga said:
Steen Bay said:
Anders W said:
I think you might be interested in the G3 ones I'll produce. I suspect you won't be happy, as they will probably show the G3 has visibly worse noise (at 100%) than the G1 at base ISO. And unlike others, I don't have faith that Oly will be able to 'tweak' the RAW files in any meaningful way (in fact, I'd be very, VERY worried that they'll do what they usually do, and sprinkle in a generous helping of nasty banding which will be in wait for those of us wanting to push and pull on the files)
Click to expand...
According to DxO the G3 beats the G1 in both (engineering) DR and SNR (18%), so it should have more 'usable DR' than the G1.
Click to expand...
I stand by what I said. And DxO agrees with me.
Click to expand...
According to DxO, the pixel-level DR is for practical purposes the same (10.06 vs 10.04) at base ISO. When the images are normalized to the same (8 MP) resolution, the G3 is slightly ahead (10.6 vs 10.3).
Click to expand...
However I didn't say anything about normalising to 8MP. I specifically said "at 100%" and at base ISO.

And under these conditions, the G1 has a SNR of 33.7 which is ahead of the G3 at 33.4db, which doesn't seem much in terms of numbers, however small differences in the SNR values are normally clearly visible to the naked eye, and this is what I see comparing G1 to G3 images. At a 100% view, the G3 files are noisier than the G1. You HAVE to downsample the G3 to get it cleaner, which is fine .... just don't consider it a 16MP sensor AND cleaner. You can have one or the other, but not both.
 
Adventsam said:
Reload your original examples in lr4, zero advantage to K-5 any more,
Click to expand...
My gut feeling tells me that this must be wrong.

Why would an improvement in LR4 ONLY affect the GH2 and not the K5 ??

The K5 sensor is better than the GH2 one. It has more latitude. If LR4 has got improved processing, I would expect it to improve the K5 sensor as well. In fact, the most likley (but of course not certain) scenario, is that if there was an IQ difference to start with, the gulf could actually WIDEN when you have the ability to push files more.

??
 
Good point --I had forgotten now that the G1 "overreads" -- by around 1/3 of a stop from what i can see and remember.

David
Anders W said:
If you base your conclusion about 1,5 stops better on your comparison between the GH2 at ISO 320 versus the G1 at 100, you need to keep in mind that the GH2 calibrates its ISO scale differently than the G1. The G1 underestimates its ISO relative to the DxO definition whereas the GH2 does not. If you put the G1 on the same ISO scale as the GH2, the G1 begins at ISO 160 rather than 100, which reduces the difference between the two points you compared to one EV. The fact that the loss of DR is typically not directly proportional to the ISO gain at the beginning of the ISO scale (due to reduced read noise when the signal is amplified), may reduce the difference you inferred to slightly less than one EV.
Click to expand...
 
You must log in or register to reply here.

Keyboard shortcuts

f
Forum
m
My threads
Mobile site
About
Editorial content
Cameras & Lenses
Community
All content, design, and layout are Copyright © 1998–2025 Digital Photography Review All Rights Reserved.
Reproduction in whole or part in any form or medium without specific written permission is prohibited.
When you use DPReview links to buy products, the site may earn a commission.
©GPS Media - Guides, Products, Services.
Back
Top