70D Dynamic Range is actually great, despite what DXOMark says

Started Jan 21, 2014 | Discussions
David Hull
David Hull Veteran Member • Posts: 6,425
Re: 70D Dynamic Range is actually great, despite what DXOMark says

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera.  It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera.  If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else).  Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings.  However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing.  My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS R Canon RF 35mm F1.8 IS STM Macro Canon RF 15-35mm F2.8L IS USM Canon RF 70-200mm F2.8L IS USM
rwbaron Forum Pro • Posts: 14,269
Re: 70D Dynamic Range is actually great, despite what DXOMark says

Just another Canon shooter wrote:

rwbaron wrote:

Just another Canon shooter wrote:

This is a really weird view of the data. You think that you can separate the read noise of the sensor from the read noise of the circuit, whatever those two mean? You are just saying that at high ISO, the two sensors look similar but we knew that.

In any case, at base ISO, the read noise is much worse.

David is absolutely correct. This has been discussed ad nauseam on the Canon forums by some very knowledgeable people. Canon's sensors are actually excellent and may be the best in the industry currently. It's their ADC being off chip that leaves the signal in analog state too long that's the problem.

How does this contradict what I said and what difference does it make? Most of the read noise comes from that circuity anyway, we know that. Amplifier noise is a major part of the "read noise" - Wikipedia. See, it is called read noise, not sensor noise. Unless we have a way to replace the ADC, we take it as a whole, and the results are known.

It does actually make a difference.  To my understanding, prior to the Sony Exmor, the ADC was always located off chip regardless of sensor manufacturer.  At that time Canon's CMOS sensors had a significant advantage and were considered the best silicon in the industry.  Enter Sony with the Exmor where they incorporated the ADC on the sensor chip where the analog to digital conversion happens immediately reducing the noise floor substantially with the benefit realized only at lower ISO's.   Canon currently has no answer for this and still uses an off chip ADC which has a much higher noise floor that penalizes their low ISO DR performance but when you consider read noise at high ISO it is evident that it's not Canon's sensor that's the issue, it's their off chip ADC design and implementation.  As David Hull has pointed out, Canon's sensors (silicon) are actually quite good if not still the best.  If Canon would implement an on chip ADC like the Exmor their overall performance might actually be the best in the industry once again.

Technically it's not the sensor, it's the ADC.

Bob

-- hide signature --
 rwbaron's gear list:rwbaron's gear list
Fujifilm X-T2 Fujifilm X-T20 Fujifilm XF 18-55mm F2.8-4 R LM OIS Fujifilm XF 55-200mm F3.5-4.8 R LM OIS Fujifilm XF 23mm F1.4 R +8 more
TTMartin
TTMartin Veteran Member • Posts: 7,304
Re: 70D Dynamic Range is actually great, despite what DXOMark says

carlk wrote:

IBIS is really not a must have for me. However I'm also at lost to explain why the companies don't want to add it in the camera even if it means they might sell a few cameras with it. The cynical explanation is they could sell more IS with each lenses but I don't think that's the reason. The unstabilized VF wouldn't be a problem for EVF os that should not be the reason either. My best guess is there are still some technical challenges to match the performance of the in lens IS for these larger sensors.

According to Nikon its not just the viewfinder.

Why is 'in-lens' VR [image stabilization] superior to 'in-camera' VR [image stabilization]?

1. Corrected finder image makes photo composition easy.
Because camera movement is compensated within the lens in use, you can see a clear finder image. This makes it easier to capture your subject in the focus frame and confirm composition. With a camera that corrects image blur inside the camera body, the image in the finder remains blurred.

2. Each lens is optimally tuned to achieve reliable correction.
Unlike cameras that only provide an internal image-blur function, every lens is optimised. As a result, you can shoot at shutter speeds several stops slower than would otherwise be possible.

3. Image information captured by the AF and metering sensors is corrected with in-lens VR.
This is a major difference from the in-camera VR. The result is faster and more accurate autofocusing and exposure metering.

4. Patterns of image blur are not the same with all lenses.
Image blur caused by camera movement differs with each lens used. This phenomenon is more noticeable when you use a lens with a longer focal length. So each lens should be finely tuned.

 TTMartin's gear list:TTMartin's gear list
Canon PowerShot S100 (2000) Canon PowerShot SD1000 Canon PowerShot G1 X Canon EOS 40D Canon EOS 400D +26 more
rwbaron Forum Pro • Posts: 14,269
Re: 70D Dynamic Range is actually great, despite what DXOMark says

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

David,

I wonder if Canon might implement a work around for this problem. From what I've seen of the early implementation of Magic Lantern's dual iso it looks very impressive.  I've not taken the time to fully understand Canon's 20 MP dual pixel sensor but possibly they could implement a better dual iso capability and still maintain full resolution but maybe I've missed something about that sensor's design.  I'd rather have the cleaner signal and extended DR at low ISO but if dual iso can be made to work as well or better for the same purpose and not sacrifice resolution then I'd be perfectly happy.  One forum member has ML's dual iso loaded in his 5D2 and has made some ISO 100 comparison's to his D600 and the 5D2/ML results look better IMO.  The downside is there's some loss of resolution but he isn't concerned because detail in shadow content isn't that obvious.

Bob

-- hide signature --
 rwbaron's gear list:rwbaron's gear list
Fujifilm X-T2 Fujifilm X-T20 Fujifilm XF 18-55mm F2.8-4 R LM OIS Fujifilm XF 55-200mm F3.5-4.8 R LM OIS Fujifilm XF 23mm F1.4 R +8 more
David Hull
David Hull Veteran Member • Posts: 6,425
Re: 70D Dynamic Range is actually great, despite what DXOMark says

rwbaron wrote:

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

David,

I wonder if Canon might implement a work around for this problem. From what I've seen of the early implementation of Magic Lantern's dual iso it looks very impressive. I've not taken the time to fully understand Canon's 20 MP dual pixel sensor but possibly they could implement a better dual iso capability and still maintain full resolution but maybe I've missed something about that sensor's design. I'd rather have the cleaner signal and extended DR at low ISO but if dual iso can be made to work as well or better for the same purpose and not sacrifice resolution then I'd be perfectly happy. One forum member has ML's dual iso loaded in his 5D2 and has made some ISO 100 comparison's to his D600 and the 5D2/ML results look better IMO. The downside is there's some loss of resolution but he isn't concerned because detail in shadow content isn't that obvious.

Bob

That seems like more of a Band-Aid approach to me.  Hopefully Canon will come up with something a bit more elegant like Sony did.  I really think that Sony was trying to address the pattern noise issue more than achieve wider DR.  I think this was probably driven by their professional broadcast clients rather than still photographers.

-- hide signature --
-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS R Canon RF 35mm F1.8 IS STM Macro Canon RF 15-35mm F2.8L IS USM Canon RF 70-200mm F2.8L IS USM
rwbaron Forum Pro • Posts: 14,269
Re: 70D Dynamic Range is actually great, despite what DXOMark says

David Hull wrote:

rwbaron wrote:

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

David,

I wonder if Canon might implement a work around for this problem. From what I've seen of the early implementation of Magic Lantern's dual iso it looks very impressive. I've not taken the time to fully understand Canon's 20 MP dual pixel sensor but possibly they could implement a better dual iso capability and still maintain full resolution but maybe I've missed something about that sensor's design. I'd rather have the cleaner signal and extended DR at low ISO but if dual iso can be made to work as well or better for the same purpose and not sacrifice resolution then I'd be perfectly happy. One forum member has ML's dual iso loaded in his 5D2 and has made some ISO 100 comparison's to his D600 and the 5D2/ML results look better IMO. The downside is there's some loss of resolution but he isn't concerned because detail in shadow content isn't that obvious.

Bob

That seems like more of a Band-Aid approach to me. Hopefully Canon will come up with something a bit more elegant like Sony did. I really think that Sony was trying to address the pattern noise issue more than achieve wider DR. I think this was probably driven by their professional broadcast clients rather than still photographers.

The ML dual ISO examples I've seen are pretty impressive and it's in it's infancy from a free developer.

Knowing large companies I've wondered if it wasn't just driven by cost reduction and component simplification. You know much more than I but it would seem a big advantage to Sony from a cost, design and layout standpoint to eliminate the off chip ADC especially when making as many as they do and the reduction in pattern noise was just a side benefit. Good selling point against the competition.

It's interesting to me that for all the hoopla over this issue on the tech related photo forums this doesn't appear to be affecting Canon, Nikon or Sony sales. Canon seems to be selling as many DSLR's as ever and appears to have even regained some market share in the pro ranks with the 1DX. It blew me away to find Photoshop RAW file guru Scott Kelby recently switch from his D4's to the 1DX. If anyone would be sensitive to this issue I would think he would be.

Bob

-- hide signature --
 rwbaron's gear list:rwbaron's gear list
Fujifilm X-T2 Fujifilm X-T20 Fujifilm XF 18-55mm F2.8-4 R LM OIS Fujifilm XF 55-200mm F3.5-4.8 R LM OIS Fujifilm XF 23mm F1.4 R +8 more
aftab
aftab Forum Pro • Posts: 10,260
Re: 70D Dynamic Range is actually great, despite what DXOMark says

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

Emil Martinec in his 2008 article shows what you two are discussing about. Read noise coming from circuitry upstream of the ISO amplifier ( I think this is what you are calling sensor read noise) and read noise coming from circuitry downstream of ISO amplifier. Canon seems to have problem with the later part. Sony and other manufacturers (Panasonic, Toshiba etc) have addressed this with on chip column parallel ADC. From memory Canon has a patent for on chip column parallel ADC since 2007. My feeling is that they haven't implemented it yet because they want to refine the technique as much as possible, add newer innovations and wait for the time when it would be cost effective.

-- hide signature --

Life is short.
Travel with passion.
http://www.flickr.com/photos/catch45/

 aftab's gear list:aftab's gear list
Canon G1 X II Panasonic FZ1000 Canon EOS 5D Canon EOS 5D Mark III Nikon D600 +25 more
David Hull
David Hull Veteran Member • Posts: 6,425
Re: 70D Dynamic Range is actually great, despite what DXOMark says

rwbaron wrote:

David Hull wrote:

rwbaron wrote:

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

David,

I wonder if Canon might implement a work around for this problem. From what I've seen of the early implementation of Magic Lantern's dual iso it looks very impressive. I've not taken the time to fully understand Canon's 20 MP dual pixel sensor but possibly they could implement a better dual iso capability and still maintain full resolution but maybe I've missed something about that sensor's design. I'd rather have the cleaner signal and extended DR at low ISO but if dual iso can be made to work as well or better for the same purpose and not sacrifice resolution then I'd be perfectly happy. One forum member has ML's dual iso loaded in his 5D2 and has made some ISO 100 comparison's to his D600 and the 5D2/ML results look better IMO. The downside is there's some loss of resolution but he isn't concerned because detail in shadow content isn't that obvious.

Bob

That seems like more of a Band-Aid approach to me. Hopefully Canon will come up with something a bit more elegant like Sony did. I really think that Sony was trying to address the pattern noise issue more than achieve wider DR. I think this was probably driven by their professional broadcast clients rather than still photographers.

The ML dual ISO examples I've seen are pretty impressive and it's in it's infancy from a free developer.

Knowing large companies I've wondered if it wasn't just driven by cost reduction and component simplification. You know much more than I but it would seem a big advantage to Sony from a cost, design and layout standpoint to eliminate the off chip ADC especially when making as many as they do and the reduction in pattern noise was just a side benefit. Good selling point against the competition.

I agree, I don't know why Canon has not abandoned their approach in favor of something different.  It may be that they save so much money on the in house sensor fab that they can afford the expensive (I assume) AFE.

It's interesting to me that for all the hoopla over this issue on the tech related photo forums this doesn't appear to be affecting Canon, Nikon or Sony sales. Canon seems to be selling as many DSLR's as ever and appears to have even regained some market share in the pro ranks with the 1DX. It blew me away to find Photoshop RAW file guru Scott Kelby recently switch from his D4's to the 1DX. If anyone would be sensitive to this issue I would think he would be.

One thing is that I think more buyers are concerned that it is clean at the higher ISO settings (which it is).  Look at Nikon releasing the specs for their new one which tops out at ISO=400k+.  What do you bet that Canon helped make it worth Scott's while to switch.

Bob

-- hide signature --
-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS R Canon RF 35mm F1.8 IS STM Macro Canon RF 15-35mm F2.8L IS USM Canon RF 70-200mm F2.8L IS USM
David Hull
David Hull Veteran Member • Posts: 6,425
Re: 70D Dynamic Range is actually great, despite what DXOMark says

aftab wrote:

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

Emil Martinec in his 2008 article shows what you two are discussing about. Read noise coming from circuitry upstream of the ISO amplifier ( I think this is what you are calling sensor read noise) and read noise coming from circuitry downstream of ISO amplifier. Canon seems to have problem with the later part. Sony and other manufacturers (Panasonic, Toshiba etc) have addressed this with on chip column parallel ADC. From memory Canon has a patent for on chip column parallel ADC since 2007. My feeling is that they haven't implemented it yet because they want to refine the technique as much as possible, add newer innovations and wait for the time when it would be cost effective.

Some have postulated that they cannot implement it within their own fab.  I suspect we will see something from Canon in the not too distant future.  I doubt that they are going to like giving up this spec to Sony for long.  Exactly what they do is the question.  I think that what keeps them going is that their current architecture is good enough since more people are concerned about clean performance at the high end.  If the press is to be believed though, the new Nikon may be applying pressure on that end now.  With an ISO rating of 400k+, 25600 is probably pretty impressive.

-- hide signature --

Life is short.
Travel with passion.
http://www.flickr.com/photos/catch45/

-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS R Canon RF 35mm F1.8 IS STM Macro Canon RF 15-35mm F2.8L IS USM Canon RF 70-200mm F2.8L IS USM
aftab
aftab Forum Pro • Posts: 10,260
Re: 70D Dynamic Range is actually great, despite what DXOMark says

David Hull wrote:

aftab wrote:

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

Emil Martinec in his 2008 article shows what you two are discussing about. Read noise coming from circuitry upstream of the ISO amplifier ( I think this is what you are calling sensor read noise) and read noise coming from circuitry downstream of ISO amplifier. Canon seems to have problem with the later part. Sony and other manufacturers (Panasonic, Toshiba etc) have addressed this with on chip column parallel ADC. From memory Canon has a patent for on chip column parallel ADC since 2007. My feeling is that they haven't implemented it yet because they want to refine the technique as much as possible, add newer innovations and wait for the time when it would be cost effective.

Some have postulated that they cannot implement it within their own fab. I suspect we will see something from Canon in the not too distant future. I doubt that they are going to like giving up this spec to Sony for long. Exactly what they do is the question. I think that what keeps them going is that their current architecture is good enough since more people are concerned about clean performance at the high end. If the press is to be believed though, the new Nikon may be applying pressure on that end now. With an ISO rating of 400k+, 25600 is probably pretty impressive.

True, more people are interested in cleaner high ISO and current Canon architecture is pretty good for that. One interesting thing I have noticed is that three best high ISO cameras in the market today have lower MP (Df, D4s and 6D). Maybe it is true that even with FF cameras smaller the pixels get you lose some high ISO performance. It would be interesting to see where the MP race goes.

As for the Magic Lantern's dual iso, I agree with you that its a band Aid type solution. But it works almost as good as the real thing. You don't get any banding. And shadows are much cleaner, so its much easier to process them. There is one extra step in processing though. And you may end up with some moire or aliasing, but I haven't encountered them most of the time to be a problem. And resolution loss is minimal (or none) if shadow is not too deep.

2 stops shadow lifting

4 stops show lifting

There are 100% crops.

 aftab's gear list:aftab's gear list
Canon G1 X II Panasonic FZ1000 Canon EOS 5D Canon EOS 5D Mark III Nikon D600 +25 more
Just another Canon shooter
Just another Canon shooter Veteran Member • Posts: 4,691
Re: 70D Dynamic Range is actually great, despite what DXOMark says

David Hull wrote:

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera.

Of course, he is modeling it. Look at the formula here. Those numbers are derived from the DXO data by curve fitting, under some assumptions (there was a recent thread about those assumptions, BTW).

 Just another Canon shooter's gear list:Just another Canon shooter's gear list
Canon EOS 5D Mark II Canon EF 15mm f/2.8 Fisheye Canon EF 35mm F1.4L USM Canon EF 50mm f/1.2L USM Canon EF 135mm F2L USM +4 more
Just another Canon shooter
Just another Canon shooter Veteran Member • Posts: 4,691
Re: 70D Dynamic Range is actually great, despite what DXOMark says

TTMartin wrote:

carlk wrote:

IBIS is really not a must have for me. However I'm also at lost to explain why the companies don't want to add it in the camera even if it means they might sell a few cameras with it. The cynical explanation is they could sell more IS with each lenses but I don't think that's the reason. The unstabilized VF wouldn't be a problem for EVF os that should not be the reason either. My best guess is there are still some technical challenges to match the performance of the in lens IS for these larger sensors.

According to Nikon its not just the viewfinder.

Why is 'in-lens' VR [image stabilization] superior to 'in-camera' VR [image stabilization]?

I think that the right question is why IBIS is superior to no stabilization at all. IBIS is not necessarily and alternative to in-lens IS, it can easily complement it for non-IS lenses.

 Just another Canon shooter's gear list:Just another Canon shooter's gear list
Canon EOS 5D Mark II Canon EF 15mm f/2.8 Fisheye Canon EF 35mm F1.4L USM Canon EF 50mm f/1.2L USM Canon EF 135mm F2L USM +4 more
David Hull
David Hull Veteran Member • Posts: 6,425
Re: 70D Dynamic Range is actually great, despite what DXOMark says

Just another Canon shooter wrote:

David Hull wrote:

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera.

Of course, he is modeling it. Look at the formula here. Those numbers are derived from the DXO data by curve fitting, under some assumptions (there was a recent thread about those assumptions, BTW).

Yes, I read that thread.  Yes, he is doing a curve fit to the DxO data.  But so what, it is the results he gets that are interesting and provide a great deal of insight into how the camera is actually working.  That is what I was really trying to point out here which I hope I have succeeded in doing.

-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS R Canon RF 35mm F1.8 IS STM Macro Canon RF 15-35mm F2.8L IS USM Canon RF 70-200mm F2.8L IS USM
TTMartin
TTMartin Veteran Member • Posts: 7,304
Re: 70D Dynamic Range is actually great, despite what DXOMark says

David Hull wrote:

aftab wrote:

David Hull wrote:

Just another Canon shooter wrote:

David Hull wrote:

So here is a question for you: As you say, the effective system noise (camera noise, if you will) is higher at ISO 100, so tell me how Canon would fix it. in other words, based on the DxO data and the Sensorgen information based on it, what part of the Canon architecture needs to be improved?

I called that read noise, not system noise, and they should improve the readout circuit. It is trivial and not worth discussing.

When you measure it, you do not know where it is coming from (but we have additional info indicating what is going on). What Sensorgen does it to model it as additive noise and a multiplicative one, and fit a curve. Then they report the additive noise as "read noise".

He isn't really modeling anything, he is taking the DxO information and presenting it in a different perspective, one which makes it a bit easier to see what is really going on in the camera. It is certainly read noise but it is not coming from the sensor, it is not trivial and it IS worth discussing for those who want to understand what is really going on in the camera. If you want to understand that DxO DR curve and why it looks the way it does, you have to take the time to understand the camera's electrical implementation, the noise line up etc.

I think one of the things that Canon grapples with is that they do have a very good sensor technology (and have had it longer than anyone else). Their chosen implementation is sub par to the competition in two areas which manifest primarily at low gain settings. However, at the other end (where probably the bulk of people use the camera) they are fine.

Some of us want to know the why behind the what, for those it probably IS worth discussing. My point was simple, the sensor is just part of the imaging system, and in the case of Canon, it does not appear to be the root cause of the low ISO deficiencies that everyone like to point out in these threads.

Emil Martinec in his 2008 article shows what you two are discussing about. Read noise coming from circuitry upstream of the ISO amplifier ( I think this is what you are calling sensor read noise) and read noise coming from circuitry downstream of ISO amplifier. Canon seems to have problem with the later part. Sony and other manufacturers (Panasonic, Toshiba etc) have addressed this with on chip column parallel ADC. From memory Canon has a patent for on chip column parallel ADC since 2007. My feeling is that they haven't implemented it yet because they want to refine the technique as much as possible, add newer innovations and wait for the time when it would be cost effective.

Some have postulated that they cannot implement it within their own fab. I suspect we will see something from Canon in the not too distant future. I doubt that they are going to like giving up this spec to Sony for long. Exactly what they do is the question. I think that what keeps them going is that their current architecture is good enough since more people are concerned about clean performance at the high end. If the press is to be believed though, the new Nikon may be applying pressure on that end now. With an ISO rating of 400k+, 25600 is probably pretty impressive.

It should be pointed out that all of this is just a theory.

If Canon's problem was truly that their 'off the shelf ADC' was too noisy so it was losing 2 bits to noise. You would think that for a camera like the 1D X Canon could simply use a 16 bit off the shelf ADC and still have 14 bits noise free.

In the Magic Lantern white paper on dual ISO they state: 'We’ll focus on the uncompressed 14-bit data found in Canon’s image buffers from the camera itself. This format is used by Canon code when saving still photos, and it is codenamed MEM1 in the firmware; some relevant strings are sdsMem1ToRawCompression and sdsMem1ToJpegDevelop.

. . .

Raw values range from some black level to some white level (usual values are 2048 and 15000) and they are linear. To convert them in stops (EV), subtract the black level and apply log2:

ev = log2(max(raw - black, 1)) (1)

where ev will be a value between 0 and almost 14. To convert from ev back to raw values:

raw = 2ev + black (2)

with some extra care to avoid going outside the [black...white] interval.

The raw image usually contains two black bars, at top and left side, which do not contain imagedata, but can be used to estimate the black level and to analyze the noise.'

Bolding added.

 TTMartin's gear list:TTMartin's gear list
Canon PowerShot S100 (2000) Canon PowerShot SD1000 Canon PowerShot G1 X Canon EOS 40D Canon EOS 400D +26 more
janfi67 Junior Member • Posts: 35
Re: 70D Dynamic Range is actually great, despite what DXOMark says

janfi67 wrote:

TTMartin wrote:

According to DxOMark every Canon dSLR they have tested basically has the same per pixel dynamic range. That includes the original 2003 APS-C Digital Rebel and the full frame 1D X. Any changes in dynamic range scores is because DxOMark uses a formula that changes the print dynamic range score based on the number of megapixels.

So you can choose to believe DxOMark, or you can look at possible flaws that would explain these seemingly erroneous results.

There is a group here that believe the results are accurate and can be explained by noise from the external analog to digital converter, I don't believe this is correct.

Oh yes, please... prove me that my lovely Canon camera is better than DxO and other say. Explain me why they are all wrong and why the DR is far better than 11EV. I can't wait to know.

Yes, keep believing that the Canon 6D, 1DX and every other Canon dSLR using the CR2 file format including the original 2003 APS C Digital Rebel have basically the same per pixel DR, because DxOMark says so, despite substantial evidence to the contrary.

I would be happy to believe you, but your argument is still rather weak. Thanks for your assertion but you didn't provide any substantial evidence.

First DxOMark attempts to measure sensor noise by looking at the RAW file. This basic concept is inaccurate because manufactures can apply different amounts noise reduction prior to the RAW file being written. Case in point the Nikon D300, Nikon D300s, and the Nikon D90 all use the same sensor. Yet the Nikon D90 scores higher than the Nikon D300 and Nikon D300s, because the D90 applies more noise reduction PRIOR to the RAW file being written.

Hummm... Like all photographers, I can only use the RAW files, not the sensor output. So I only care about what the RAW file contains and what I can do with it.

And as said previously, you're wrong. D300 and D90 don't use the same sensor.

Then you should care that Canon Digital Photo Professional (DPP) does a better job at decoding the RAW file, and not what testing sites using a 3rd party RAW converter say.

That's your opinion, and I respect it, even if mine is different. By the way, I don't need testing sites to know which RAW converter I prefer. You'll notice the subtle difference, I dot not affirm one is better than the others, I'm just saying I prefer one to the others.

Camera that use Sony sensors apply noise reduction at the sensor level prior to the RAW file being written.

Do you think Canon doesn't apply some kind of noise reduction prior the RAW file being written? Never heard of CDS? It's embedded in the sensor itself (including the Canon ones) and it's a noise cancellation system.

If all manufactures apply some noise reduction prior to the RAW file being written then what exactly is DxOMark testing? Certainly not the sensor performance like they claim.

I do not care about what DxOMark is testing. I just highlighted that Canon do the same kind of things than Sony in this area.

Canon embeds information in the RAW file to be used by their RAW converter DPP. Third party RAW converters don't take advantage of this information. For example Canon masks portions of the edge of their sensor, to provide information on both row and column sensor noise, 3rd party RAW converters like the one used by DxOMark do not use this information.

You seems misinformed. Take a look at public domain RAW converters sources and documentation. After reverse engineering they use data provided by the masked portions of the edge of the sensor to reduce the noise.

When I looked at DCRAW, it did not. If I am wrong please post the portion of code that uses it.

That's easy, hare are some extracts of an old, but easy to understand version of Dcraw sources:

unsigned black, cblack[8];

void CLASS lossless_jpeg_load_raw()
{
...
for (jrow=0; jrow < jh.high; jrow++) {
rp = ljpeg_row (jrow, &jh);
for (jcol=0; jcol < jwide; jcol++) {
val = *rp++;
...
if ((unsigned) (row-top_margin) < height) {
c = FC(row-top_margin,col-left_margin);
if ((unsigned) (col-left_margin) < width) {
BAYER(row-top_margin,col-left_margin) = val;
if (min > val) min = val;
} else if (col > 1 && (unsigned) (col-left_margin+2) > width+3) // end of the row
cblack[c] += (cblack[4+c]++,val);
...
} // end of for (jcol
} // end of for (jrow
ljpeg_end (&jh);
FORC4 if (cblack[4+c]) cblack[c] /= cblack[4+c];
...
}

As you can see, cblack is computed using values of the masked portions of the sensor.

In the latest version it is located in the crop_masked_pixels class and it uses mblack array.

And how can you tell than products like DxO or ACR don't use them?

Because their conversions contain more noise than those converted with Canon DPP.

No, seriously, I was waiting for a more elaborate answer, not for this rather weak reasoning.

Are you aware of all the details about CR2 format Canon discloses only under NDA?

Are you? Can you name someone other than a Canon employee who is?

Once again, you're not answering to the question.

By reverse engineering, at least one public domain RAW converter use some proprietary information of the RAW file . Don't you think Adobe or DxO people cannot do at least the same?

And anyone working on reverse engineering the CR2 file would be in violation of their NDA, so any implication that you or anyone else has more knowledge about what Canon actually does in the CR2 file is just a red herring.

As usual, you don't understand what you are writing. Do you know what a NDA is? Not obviously. A Non Disclosure Agreement engage you to not disclose information given by the owner of this information. In case of reverse engineering, there is no NDA! You may in some cases violate intellectual properties laws, but obviously not a NDA that you have not signed.

Because their conversions contain more noise than those converted with Canon DPP.

Always the same assertion. That's rather weak.

Third party RAW converters attempt to directly read portions of the Canon CR2 file and directly convert them to RGB. There are several problems with this, first according to Canon's Chuck Westfall Canon RAW data is recorded in sYCC and not RGB.

This is one of the funniest part of your post, and once again, you're wrong.

Do you realize what does recording RAW data in sYCC mean? I don't think so, despite the picture you post.

An sYCC encoded image is obtained AFTER DEMOSAICING the CFA RGB data. It's not a RAW format.

This is where your thinking falls apart. You assume because the original Canon CRW format and Nikon NEF used a certain format to encode their RAW data that Canon had to be constrained by that when they developed the CR2 format. A RAW format is what the manufacturer wants to use to maintain all the data from the sensor. Sorry, the manufacturer does not have to be constrained by your definition of how sensor data is maintained in the RAW file.

I'm sorry, but your answer is totally off topic and you're wrong. Canon DSLR RAW data for the full size raw are CFA RGGB data, not sYCC data. You don't believe me? Read Understanding What is stored in a Canon RAW .CR2 file, How and Why. If you don't know where to find this information, it is §3.4.4. Decoding image data :"The output of the jpeg lossless decompressing is a Color Filter Array (CFA) picture."

Switch a Nikon D300 to 14 bit NEF and it slows to 3 fps, the Nikon D7100 isn't much better.

So what? What is the relation with Nikon? We're talking about how are encoded Canon CR2 files.

So you don't think it's possible that when Canon developed the CR2 file format that they recognized that in order to move 14 bit RAW data more quickly that they had to find a different way to encode it?

try to understand Understanding What is stored in a Canon RAW .CR2 file, How and Why.

For your information, Canon sRAW formats use YCC encoding, but they aren't RAW format per se, and they are not used by DxO to perform their tests. It's a kind of lossless jpeg, but using 15 bits data for the 3 components, Y, Cr and Cb, not 8 bits like traditional jpeg.

sRAW is not a RAW format because you say so? Canon says it is a RAW format. Again a RAW format is whatever format the manufacturer wants to use to maintain all the necessary data from the sensor.

I should have guessed you don't understand latin phrase, sorry for that. But no, stricto sensu, sRAW is not a RAW format because it has been democaised. Once democaised, some information from the sensor are lost. But who cares about wording?

So the process of direct conversion to RGB often result in unexpected results, like this CR2 image from my Canon 6D with a 3rd party RAW converter that doesn't know the 6D exists.

It seems you have no idea of what the CR2 format is, and how it differs for 6D compared to previous Canon cameras. There are many differences that can explain a RAW converter that doesn't know the 6D exists fails to convert a 6D image.

For example, since Canon 50D and digic IV, CFA RAW data are encoded in 4 components into the CR2 files. The 6D (and the 1Dx) use only 2 components. Tricky for a RAW converter, isn't it? This has nothing to do with sYCC.

No, it is an example of how the CR2 format hasn't been reverse engineered. Each camera that comes out, 3rd party RAW converters have to go in and look for what data that they can find that matches their concept of what RAW data should look like. And then force that round data into a square hole.

Sorry, but if you 3rd party RAW converter was not able do decode your 6D RAW file, it was not because it was encoded in sYCC. You should now know that it contains RGGB CFA data. By the way, you underestimate a lot the power of the reverse engineering.

Another flaw in this attempted direct conversion to RGB is that it uses the HSB/HSV color space instead of the HSL color space used by Canon.

Once again, you don't understand what you're talking about. HSL is used in Canon picture style editor. It has nothing to do with demosaicing.

Any time you demosaic it has to be done into a color space. And this is just another clue that Canon isn't simply using RGB to encode the data in the CR2 file. 3rd party RAW converters demosaic using the assumption that the data is in an HSB color space like the CRW and NEF formats use. It's clear from the Canon Picture Styles that Canon is using the HSL color space in camera (Picture Styles affect in camera JPG processing). Canon writing the RAW data into the HSL color space would explain why Canon is so much more efficient in moving RAW data.

Writing a lot a time the same wrong assertion won't make it true. Hopefully, 3rd party RAW converters don't use the assumption that the CR2 data are in HSL color space. Otherwise, they wouldn't be able to provide any image.

So what does all this actually mean. In my opinion the Canon CR2 file has been incorrectly reverse engineered by 3rd party RAW converters. This incorrect reverse engineering has only resulted in 12 bits of usable data per pixel and not the full 14 bits. And it is this that actually explains why testing sites that use 3rd party RAW converters show that Canon's per pixel dynamic range has remained unchanged in the last 10 years and not that the 2003 Digital Rebel and the Canon 1D X have the same per pixel dynamic range.

I'm really disappointed. I expected strong evidences, solid technical arguments. But no, it was only blabla and technical words ans concepts not understood by a Canon fanboy.

Ditto, except what you be called a DxOMark fanboy?

Did I wrote somewhere that DxO is good or bad?

At least I do not bash DxOMark when I don't like their results about DR and use their measures when I agree with them that more than 20Mp is useless: http://www.dpreview.com/forums/post/53280449

By the way. I'm very happy with my Canon cameras even if they have less DR than other brands.

Maybe that's because they really don't.

TTMartin
TTMartin Veteran Member • Posts: 7,304
Re: 70D Dynamic Range is actually great, despite what DXOMark says

janfi67 wrote:

janfi67 wrote:

TTMartin wrote:

Are you aware of all the details about CR2 format Canon discloses only under NDA?

Are you? Can you name someone other than a Canon employee who is?

Once again, you're not answering to the question.

Nor have you.

By reverse engineering, at least one public domain RAW converter use some proprietary information of the RAW file . Don't you think Adobe or DxO people cannot do at least the same?

And anyone working on reverse engineering the CR2 file would be in violation of their NDA, so any implication that you or anyone else has more knowledge about what Canon actually does in the CR2 file is just a red herring.

As usual, you don't understand what you are writing. Do you know what a NDA is? Not obviously. A Non Disclosure Agreement engage you to not disclose information given by the owner of this information. In case of reverse engineering, there is no NDA! You may in some cases violate intellectual properties laws, but obviously not a NDA that you have not signed.

I understand perfectly well what an NDA is.

If someone had signed an NDA with Canon.

They would be in violation of their NDA with Canon if they worked with a 3rd party team to reverse engineer the CR2 file.

So asking if I have an NDA with Canon is a red herring, as no one working to reverse engineer the CR2 file should have an NDA with Canon.

 TTMartin's gear list:TTMartin's gear list
Canon PowerShot S100 (2000) Canon PowerShot SD1000 Canon PowerShot G1 X Canon EOS 40D Canon EOS 400D +26 more
Keyboard shortcuts:
FForum MMy threads