Testing my GH2's sensor

Started Feb 20, 2011 | Discussions
EXR
EXR
Senior MemberPosts: 1,437
Like?
Re: Testing my GH2's sensor
In reply to Kolen, Apr 22, 2011

Kolen wrote:

Hi, from your result, only 160, 320, 640, etc. are native ISOs. I am thinking if DxOmark will then underestimated the sensor performance since they measure the performance at 160, 200, 400, etc.

This shouldn´t make any difference. DXO measures noise, or to be more accurate signal-noise-ratio.

Amplifying a signal will of course decrease SNR, but it shouldn´t matter if you do it analogue or digital. It is rather the opposite, each analogue device will potentially increase noise, so an analogue amplifier could increase noise.

A digital multiplication (which is the equivalent of an amplification) will never increase noise.

When we analyse the DXO-data a bit closer beyond just reading the numbers in front we can see a few things.

Generally the GH1 is better per pixel, which is to be expected with fewer pixels on the same area.

Resolution-normalized both sensors show almost exactly the same performance in most areas.

But there are some interesting things to see.

An ideal sensor should show a 3dB difference of SNR when you increase/decrease ISO by 1 stop.

When we look at the GH1 sensor it matches almost perfectly with the theory, with one exception. ISO 800 is better by almost 1dB (equivalent to 1/3 stop) then expected. But this path doesn´t continue, ISO 1600 and 3200 are about in line with the other ISOs. Looking at the full SNR-data we see that the brightest areas show indeed the expected 3dB difference between ISO400 and 800. The 2 curves get closer at lower grey-values.

This would indicate lower read-noise for ISO800, although there is no plausible explanation why this should happen suddenly and only at ISO800. (and the curves start to get closer at 50% grey, which is a bit to much to be just read-noise)

Anyway, this "extraordinary" ISO800-performance is mainly responsible for the "victory" of the GH1 in the low-light area.

Lets take a look at the GH2.

The first thing we can see is, that we have just 2,7dB difference between ISO200 and 400, a bit lower than expected. This indicates somewhat higher read-noise for the GH2.

Looking at the graphs for the Full SNR-data of both cameras we can see the lines of the GH1 stay straight a bit longer in the area below 1% Grey.

This backs up our theory, that the GH2 indeed has a very slightly, but measurable, higher readout-noise throughout the ISO-range. (possibly a sight-effect of the higher readout-speed)

We also see the GH2 is missing the lowest ISO, starting at 160 instead of 100 (130 in reality). This is mainly responsible for the higher DR-score of the GH1.

We also see a slightly disappointing performance of the GH2 at the highest ISOs.

Between ISO 6400 and 12800 it loses almost 6dB in SNR, which is almost twice as much as expected. I don´t know any useful explanation for this behavior.

In the end the 2 sensors seem to use very similar technology and the difference seems to be very small. Maybe the just had a "good" sample of a GH1 and a rather "bad" sample of the GH2 for the test.

The pure sensor-quality seems to be more or less equivalent between those 2 cameras.

Reply   Reply with quote   Complain
Kolen
Senior MemberPosts: 1,015Gear list
Like?
Re: Testing my GH2's sensor
In reply to EXR, Apr 22, 2011

EXR wrote:

Kolen wrote:

Hi, from your result, only 160, 320, 640, etc. are native ISOs. I am thinking if DxOmark will then underestimated the sensor performance since they measure the performance at 160, 200, 400, etc.

This shouldn´t make any difference. DXO measures noise, or to be more accurate signal-noise-ratio.

Amplifying a signal will of course decrease SNR, but it shouldn´t matter if you do it analogue or digital. It is rather the opposite, each analogue device will potentially increase noise, so an analogue amplifier could increase noise.

A digital multiplication (which is the equivalent of an amplification) will never increase noise.

When we analyse the DXO-data a bit closer beyond just reading the numbers in front we can see a few things.

Generally the GH1 is better per pixel, which is to be expected with fewer pixels on the same area.

Resolution-normalized both sensors show almost exactly the same performance in most areas.

But there are some interesting things to see.

An ideal sensor should show a 3dB difference of SNR when you increase/decrease ISO by 1 stop.

When we look at the GH1 sensor it matches almost perfectly with the theory, with one exception. ISO 800 is better by almost 1dB (equivalent to 1/3 stop) then expected. But this path doesn´t continue, ISO 1600 and 3200 are about in line with the other ISOs. Looking at the full SNR-data we see that the brightest areas show indeed the expected 3dB difference between ISO400 and 800. The 2 curves get closer at lower grey-values.

This would indicate lower read-noise for ISO800, although there is no plausible explanation why this should happen suddenly and only at ISO800. (and the curves start to get closer at 50% grey, which is a bit to much to be just read-noise)

Anyway, this "extraordinary" ISO800-performance is mainly responsible for the "victory" of the GH1 in the low-light area.

Lets take a look at the GH2.

The first thing we can see is, that we have just 2,7dB difference between ISO200 and 400, a bit lower than expected. This indicates somewhat higher read-noise for the GH2.

Looking at the graphs for the Full SNR-data of both cameras we can see the lines of the GH1 stay straight a bit longer in the area below 1% Grey.

This backs up our theory, that the GH2 indeed has a very slightly, but measurable, higher readout-noise throughout the ISO-range. (possibly a sight-effect of the higher readout-speed)

We also see the GH2 is missing the lowest ISO, starting at 160 instead of 100 (130 in reality). This is mainly responsible for the higher DR-score of the GH1.

We also see a slightly disappointing performance of the GH2 at the highest ISOs.

Between ISO 6400 and 12800 it loses almost 6dB in SNR, which is almost twice as much as expected. I don´t know any useful explanation for this behavior.

In the end the 2 sensors seem to use very similar technology and the difference seems to be very small. Maybe the just had a "good" sample of a GH1 and a rather "bad" sample of the GH2 for the test.

The pure sensor-quality seems to be more or less equivalent between those 2 cameras.

Thanks for interpreting the data. It is clearer. I think at least GH2 has much better control on banding noise then. This alone can compensate to any slightly lower score in the DxOMark.

I heard that Topaz Denoise has banding removal. I think I should try it out soon.

By the way, I think I still don't understand it completely. jumping from ISO 600 to 12800 shows that its digital pull is not perfect. Actually Canon is infamous for that too so users are warned by the others not to use ISO 125, 250, etc. I think a video posts somewhere in this post shows similar behavior for GH1/GH2. So that could mean using the ISO200, 400, etc. to measure the noise in DxOMark might underestimate the sensor performance at native ISOs.
I might be wrong, though.

 Kolen's gear list:Kolen's gear list
Canon EOS 5D Mark II Sony Alpha NEX-C3 Nikon D800E Olympus PEN E-PL5 Canon EF 200mm f/2.8L II USM +32 more
Reply   Reply with quote   Complain
kenw
Veteran MemberPosts: 4,486Gear list
Like?
Re: Testing my GH2's sensor
In reply to Kolen, Apr 22, 2011

Kolen wrote:

By the way, I think I still don't understand it completely. jumping from ISO 600 to 12800 shows that its digital pull is not perfect.

Yeah, I can't figure out what is going on there. At first I thought it might be because by 12800 the SNR 18% curves were becoming read noise instead of photon shot noise limited. I just looked at the full SNR curves though and that doesn't appear to be the case, the entire SNR curve is about 6 dB down starting from saturation - makes no sense to me. I really don't understand what is going on at 12800 - either with the camera or DxO's measurement. If the DxO data is valid one should never use ISO 12800 and should just push ISO 6400 instead. If I have sometime maybe I'll do a test.

Actually Canon is infamous for that too so users are warned by the others not to use ISO 125, 250, etc.

So Canon had done some things differently. They weren't underexposing and multiplying, they were overexposing and dividing. This leads to lower dynamic range. A number of their cameras also have an unusual SNR behavior at lower ISOs. If I recall it is on account of two different read noise sources (before and after the VGA I think). That is also probably making the Canon case a bit different.

I think a video posts somewhere in this post shows similar behavior for GH1/GH2. So that could mean using the ISO200, 400, etc. to measure the noise in DxOMark might underestimate the sensor performance at native ISOs.
I might be wrong, though.

Possible, but the one "native" ISO they did measure (160) is exactly on the trend line for the 18% SNR and actually dips a bit on DR (which it probably should as things are getting read noise limited in that region). Looking at the full SNR curves the "fake" ISOs are behaving exactly as expected compared to 160, no sign of trouble. The outlier is 12800. I don't know what is going on there, whether it is the camera or something with DxO's measurement.

-- hide signature --

Ken W
See plan in profile for equipment list

 kenw's gear list:kenw's gear list
Sony RX100 Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Panasonic Lumix DMC-GM1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS +29 more
Reply   Reply with quote   Complain
EXR
EXR
Senior MemberPosts: 1,437
Like?
Re: Testing my GH2's sensor
In reply to Kolen, Apr 22, 2011

Kolen wrote:

Thanks for interpreting the data. It is clearer. I think at least GH2 has much better control on banding noise then. This alone can compensate to any slightly lower score in the DxOMark.

I can´t see any banding anymore with my GH1 with the latest firmware and ACR6 or LR3.

Again the difference is so small it´s not really worth thinking much about it and as I it comes mostly from a "too good" ISO800 of the GH1. So it could be a wrong measurement (although I believe DXO is professional enough to verify its measurements, especially if they are rather strange)

By the way, I think I still don't understand it completely. jumping from ISO 600 to 12800 shows that its digital pull is not perfect. Actually Canon is infamous for that too so users are warned by the others not to use ISO 125, 250, etc. I think a video posts somewhere in this post shows similar behavior for GH1/GH2. So that could mean using the ISO200, 400, etc. to measure the noise in DxOMark might underestimate the sensor performance at native ISOs.

A Multiplication by 2 should be the same digital or analogue.

Of course you lose precision when doing it digital, but this shouldn´t make any difference when the low order bits only contain random data due to noise, and more important shouldn´t increase noise.

Maybe something in the measurement-method of DXO is indeed somehow affected by the "fake" ISOs, I wouldn´t rule out this possibility completely.

Reply   Reply with quote   Complain
Kolen
Senior MemberPosts: 1,015Gear list
Like?
Re: Testing my GH2's sensor
In reply to kenw, Apr 22, 2011

kenw wrote:

Kolen wrote:

By the way, I think I still don't understand it completely. jumping from ISO 600 to 12800 shows that its digital pull is not perfect.

Yeah, I can't figure out what is going on there. At first I thought it might be because by 12800 the SNR 18% curves were becoming read noise instead of photon shot noise limited. I just looked at the full SNR curves though and that doesn't appear to be the case, the entire SNR curve is about 6 dB down starting from saturation - makes no sense to me. I really don't understand what is going on at 12800 - either with the camera or DxO's measurement. If the DxO data is valid one should never use ISO 12800 and should just push ISO 6400 instead. If I have sometime maybe I'll do a test.

Actually Canon is infamous for that too so users are warned by the others not to use ISO 125, 250, etc.

So Canon had done some things differently. They weren't underexposing and multiplying, they were overexposing and dividing. This leads to lower dynamic range. A number of their cameras also have an unusual SNR behavior at lower ISOs. If I recall it is on account of two different read noise sources (before and after the VGA I think). That is also probably making the Canon case a bit different.

I think a video posts somewhere in this post shows similar behavior for GH1/GH2. So that could mean using the ISO200, 400, etc. to measure the noise in DxOMark might underestimate the sensor performance at native ISOs.
I might be wrong, though.

Possible, but the one "native" ISO they did measure (160) is exactly on the trend line for the 18% SNR and actually dips a bit on DR (which it probably should as things are getting read noise limited in that region). Looking at the full SNR curves the "fake" ISOs are behaving exactly as expected compared to 160, no sign of trouble. The outlier is 12800. I don't know what is going on there, whether it is the camera or something with DxO's measurement.

-- hide signature --

Ken W
See plan in profile for equipment list

Thanks. I think Canon's ISO is this: ISO125, 250, etc. is pushed from ISO100 and 200, etc. And ISO50, 160, 320, etc. are pulled from ISO100, 200, 400, etc. But I am not sure exactly why Canon shows more noise at ISO125, 250, etc. I think it is because Canon's read noise is quite high, showing up a lot in low ISOs, so pushing at low ISO like ISO100 to 125 results in even more noise than ISO200.
There is a video posted a while ago in this post: http://vimeo.com/18580410

It shows that ISO 250, 500, 1000 and 2000 has more apparent noise. It actually behaves very Canon like that 125, 250, 500, etc. should be avoided. And this tempt me to conclude that 250 from GH2 is pushed from 200, which is not true according to the original post in the very beginning.
Anyway, there is so many mystery.

 Kolen's gear list:Kolen's gear list
Canon EOS 5D Mark II Sony Alpha NEX-C3 Nikon D800E Olympus PEN E-PL5 Canon EF 200mm f/2.8L II USM +32 more
Reply   Reply with quote   Complain
Kolen
Senior MemberPosts: 1,015Gear list
Like?
Re: Testing my GH2's sensor
In reply to EXR, Apr 22, 2011

EXR wrote:

Kolen wrote:

Thanks for interpreting the data. It is clearer. I think at least GH2 has much better control on banding noise then. This alone can compensate to any slightly lower score in the DxOMark.

I can´t see any banding anymore with my GH1 with the latest firmware and ACR6 or LR3.

Oh, really? I just received an GH1 a few days ago and has banding. I thought that was normal. Should I return it to Panasonic.com?

Again the difference is so small it´s not really worth thinking much about it and as I it comes mostly from a "too good" ISO800 of the GH1. So it could be a wrong measurement (although I believe DXO is professional enough to verify its measurements, especially if they are rather strange)

By the way, I think I still don't understand it completely. jumping from ISO 600 to 12800 shows that its digital pull is not perfect. Actually Canon is infamous for that too so users are warned by the others not to use ISO 125, 250, etc. I think a video posts somewhere in this post shows similar behavior for GH1/GH2. So that could mean using the ISO200, 400, etc. to measure the noise in DxOMark might underestimate the sensor performance at native ISOs.

A Multiplication by 2 should be the same digital or analogue.

Of course you lose precision when doing it digital, but this shouldn´t make any difference when the low order bits only contain random data due to noise, and more important shouldn´t increase noise.

Maybe something in the measurement-method of DXO is indeed somehow affected by the "fake" ISOs, I wouldn´t rule out this possibility completely.

 Kolen's gear list:Kolen's gear list
Canon EOS 5D Mark II Sony Alpha NEX-C3 Nikon D800E Olympus PEN E-PL5 Canon EF 200mm f/2.8L II USM +32 more
Reply   Reply with quote   Complain
Corpy2
Senior MemberPosts: 1,417
Like?
Re: Testing my GH2's sensor
In reply to bg2b, Aug 5, 2011

You say there's no point in going past 640 or whatever, but then how do I deal wit low light and stop motion issues?

I need high ISO settings in order to get exposed shots at reasonable shutter speeds.

Reply   Reply with quote   Complain
Corpy2
Senior MemberPosts: 1,417
Like?
Re: Testing my GH2's sensor
In reply to bg2b, Aug 5, 2011

Why does an artificially pushed ISO histogram show those empty spaces?

Reply   Reply with quote   Complain
jimboyvr
Senior MemberPosts: 1,216
Like?
Re: Testing my GH2's sensor
In reply to cameron2, Aug 5, 2011

cameron2 wrote:

So what do you set the ISO at when it's really dark and/or you need to freeze motion? i.e. if ISO 3200 is called for, do you still shoot at 640 and just fix it in PP?

was this answered. excellent question.

Reply   Reply with quote   Complain
Corpy2
Senior MemberPosts: 1,417
Like?
Re: Testing my GH2's sensor
In reply to jimboyvr, Aug 5, 2011

Yes, it was asked a few times and answered. I had not seen the questions before I posted mine. See earlier responses.

The basic answer is to dial in underexposure and use the lower ISO speed, which can't hurt in comparison with simply shooting at the higher speed and may help with highlights.

Reply   Reply with quote   Complain
Keyboard shortcuts:
FForum MMy threads