What happened to Canon sensor development?

rwbaron · Apr 6, 2012

Interesting results using DXO's software and procedure to measure DR and SNR.

Now we need the debate as to why this doesn't correspond with the 2+ stops of DR the D800 is supposed to have over the 5D3. At ISO 100 to about 700 the Tiff's are showing less than a 1/2 stop advantage for the D800 and past 800 the 5D3 is better by about the same value. Furthermore, the chart shows the Tiff SNR of the 5D3 to be higher than the D800 across all ISO's except for 800. Are these people that inept or is there something the techies on this forum have missed?

I'm very interested in the explanation.

Bob

Mr.Toad said:
Here:

http://www.techradar.com/news/photography-video-capture/cameras/hands-on-nikon-d800-review-1061286/page:5#articleContent

To see the chart legend you have to right click and select "View Image" (firefox - in IE, I had to "save picture as").
--
http://www.pbase.com/mrtoad

--
http://www.pbase.com/rwbaron

David Hull · Apr 6, 2012

Anastigmat said:
carlk wrote:

You have no idea what you are talking about, with all due respect. Canon holds many patents on sensor technology that Sony and Fuji have no access to. It is Sony and Fuji who wish they have Canon's on chip per pixel noise reduction technology, not vice versa.

I don’t know about Carl, but you are a bit mixed up on the facts as they really stand. Sony already has on-chip, per-pixel noise reduction technology that is all digital and very effective. Any patents that Canon has WRT the CDS (which is what you are talking about) are of no use to Sony who has already rejected the analog approach taken by Canon. Unfortunately, it is Canon who apparently lacks access to key Sony/IBM patents. Sony's on-chip, per-pixel noise reduction technology has given them very low, very non-deterministic read noise which Canon unfortunately cannot match with their current analog approach.

That being said, this appears to be about the only thing lacking for Canon, they seem to have the other end of the ISO range working well and read noise is not the only parameter of concern in the design of sensors. The jury (AFAIK) is still out on other things like Qe that help that end. They might have made improvements there that we don't know about yet. Unfortunately there is not nearly as much room for improvement on that end -- everyone was doing pretty well.

--
Flickr: http://www.flickr.com/photos/26158506@N07/

Horshack · Apr 6, 2012

rwbaron said:
Interesting results using DXO's software and procedure to measure DR and SNR.

Now we need the debate as to why this doesn't correspond with the 2+ stops of DR the D800 is supposed to have over the 5D3. At ISO 100 to about 700 the Tiff's are showing less than a 1/2 stop advantage for the D800 and past 800 the 5D3 is better by about the same value. Furthermore, the chart shows the Tiff SNR of the 5D3 to be higher than the D800 across all ISO's except for 800. Are these people that inept or is there something the techies on this forum have missed?

I'm very interested in the explanation.

Those aren't DxoMark's results. DxoMark's testing methodology requires precise control over the lighting setup and they don't test TIFFs.

altair8800 · Apr 6, 2012

carlk said:
On the other hand for whatever reasons this (same pixel pitch for full frame as aps-c) has never been done before.

When the 1DsIII came out, it had the same pixel size as the 20D. I had a 20D at that time and considered the 1DsIII as I could have the same pixel density for birding and have a much larger area for cropping and better AF. But it soon lost the lead in pixel density.

carlk said:
Sony already have 24MP aps-c which is equivalent to 54MP FF but they only made the latest one 36MP. I’m not sure if Canon can do a 46MP even if they are willing to but we can always hope.

I am sure they can do 46Mp, but producing it at a price to compete with the D800 may be a problem. They would have to put in dual Digic 5 processors like the 1DX to get a reasonable frame rate. Maybe three Digic 5?

Dan

rwbaron · Apr 6, 2012

Horshack said:
rwbaron said:

Interesting results using DXO's software and procedure to measure DR and SNR.

Now we need the debate as to why this doesn't correspond with the 2+ stops of DR the D800 is supposed to have over the 5D3. At ISO 100 to about 700 the Tiff's are showing less than a 1/2 stop advantage for the D800 and past 800 the 5D3 is better by about the same value. Furthermore, the chart shows the Tiff SNR of the 5D3 to be higher than the D800 across all ISO's except for 800. Are these people that inept or is there something the techies on this forum have missed?

I'm very interested in the explanation.

Click to expand...

Those aren't DxoMark's results. DxoMark's testing methodology requires precise control over the lighting setup and they don't test TIFFs.

Did you read the article and click on the DR and SNR measurement explanation? They use a backlit target source designed specifically for this type of measurement which they say is more accurate. They use Tiff's because Dxo's software will not read the RAW files at this time. I know how to work with a Tiff but not DxoMark's results to print an image but maybe you can explain. Maybe I'm missing something but this test seems to approximate the real World more than DxoMark's results.

Bob
--
http://www.pbase.com/rwbaron

Horshack · Apr 6, 2012

rwbaron said:
Did you read the article and click on the DR and SNR measurement explanation? They use a backlit target source designed specifically for this type of measurement which they say is more accurate. They use Tiff's because Dxo's software will not read the RAW files at this time. I know how to work with a Tiff but not DxoMark's results to print an image but maybe you can explain. Maybe I'm missing something but this test seems to approximate the real World more than DxoMark's results.

http://www.dxomark.com/index.php/About/In-depth-measurements/DxOMark-testing-protocols/Noise-dynamic-range

bobn2 · Apr 6, 2012

Anastigmat said:
You have no idea what you are talking about, with all due respect.

With all due respect, you certainly have no idea what you are talking about, despite having your nonsense put straight many times.

Anastigmat said:
Canon holds many patents on sensor technology that Sony and Fuji have no access to. It is Sony and Fuji who wish they have Canon's on chip per pixel noise reduction technology, not vice versa.

Canon's 'on chip per pixel noise reduction technology' is simply correlated double sampling (I remember having told you this, time and time again, most recently here
http://forums.dpreview.com/forums/read.asp?forum=1032&message=40801354

yet still you go on posting about the 'Canon on chip per pixel noise reduction technology'), and the classic 4-T pixel circuit to allow it. This configuration is used already by Sony, Nikon, Toshiba (which makes 'Fuji' sensors), Aptina, Omnivision, Samsung, uncle Tom Cobbley and all. Canon's 4-T configuration is not substantially different from the others, so there is nowhere 'per-pixel' for any different 'noise reduction' mechanism to reside. It is not patentable, having been originally patented by Eric Fossum's team at Caltech in 1994 (US5841126) - thus out of force.

What I believe Canon has patented (though it is tough finding what it is searching Canon's thousands of patents) is a front end configuration on-chip (per column line, not per pixel) to collect the two samples. It is not clear that other solutions do not work as well, and the (also patented) mechanism built into the column ADC's on Sony's Exmor sensor probably works better. One good reason to suspect that it is a better solution is that the Sony sensors are considerably freer of pattern noise, which is one of the things CDS is trying to eliminate. So, since they have a better solution, it is highly unlikely that Sony is coveting Canon's.

--
Bob

rwbaron · Apr 6, 2012

http://www.techradar.com/news/photography-video-capture/cameras/noise-and-dynamic-range-results-explained-1027588

Where's the difference?

Horshack said:
rwbaron said:

Did you read the article and click on the DR and SNR measurement explanation? They use a backlit target source designed specifically for this type of measurement which they say is more accurate. They use Tiff's because Dxo's software will not read the RAW files at this time. I know how to work with a Tiff but not DxoMark's results to print an image but maybe you can explain. Maybe I'm missing something but this test seems to approximate the real World more than DxoMark's results.

Click to expand...

http://www.dxomark.com/index.php/About/In-depth-measurements/DxOMark-testing-protocols/Noise-dynamic-range

--
http://www.pbase.com/rwbaron

kikiriki · Apr 6, 2012

Can they do high MP sensor? Yes, they did it..

http://www.dpreview.com/news/2010/8/24/canon120mpsensor

It is just matter of timing, and some people will never understand that..

In business timing is everything, and Canon team for some reason think it is better to see what competition can do, and then answer with clear targets..
--
Igor Leskovar

bobn2 · Apr 6, 2012

Posted in haste, sorry.
--
Bob

Mr.Toad · Apr 6, 2012

bobn2 said:
not DxO (using DxO tools)

True.

bobn2 said:
Those are not the 5DIII

Untrue. The yellow plot is the 5D3.

bobn2 said:
Ignore this

Ah - ok ;-)

--
http://www.pbase.com/mrtoad

Daniel Clune · Apr 6, 2012

Well if canon needs a higher meg camera to keep people leaving canon and going else were they need it now. So if timing is important then if they wait till sept 2012 as some have guessed at thats a little to late for anyone that is considering going to nikon D800. Plus my bet by mid summer Sony will also have there 36 meg camera out by then. So timing is important and if and thats a BIG if canon has something soon for at least 2012 they should announce it now. Wait till sept or later and anyone that wanted to go nikon will have done so by then.

kikiriki said:
Can they do high MP sensor? Yes, they did it..

http://www.dpreview.com/news/2010/8/24/canon120mpsensor

It is just matter of timing, and some people will never understand that..

In business timing is everything, and Canon team for some reason think it is better to see what competition can do, and then answer with clear targets..
--
Igor Leskovar

--
http://www.pbase.com/dc9mm

oysso · Apr 6, 2012

kikiriki said:
Can they do high MP sensor? Yes, they did it..

http://www.dpreview.com/news/2010/8/24/canon120mpsensor

It is just matter of timing, and some people will never understand that..

In business timing is everything, and Canon team for some reason think it is better to see what competition can do, and then answer with clear targets..
--
Igor Leskovar

I bet someone do. But many don't want to understand because they want that High MP camera NOW.

bobn2 · Apr 6, 2012

rwbaron said:
Interesting results using DXO's software and procedure to measure DR and SNR.

Now we need the debate as to why this doesn't correspond with the 2+ stops of DR the D800 is supposed to have over the 5D3. At ISO 100 to about 700 the Tiff's are showing less than a 1/2 stop advantage for the D800 and past 800 the 5D3 is better by about the same value. Furthermore, the chart shows the Tiff SNR of the 5D3 to be higher than the D800 across all ISO's except for 800. Are these people that inept or is there something the techies on this forum have missed?

I'm very interested in the explanation.

Those are very strange curves. Compare their curve

with DxO's DR curve for the D800
100: TR has 12.7, DxO 13.23
200: 12.6, 12.56
400, 12.6, 11.74
800, 11.7, 10.93
1600, 11.7, 10.03
3200, 10.2, 9.0
6400, 9.3, 8.07
12800, 8.7, 7.06

So, while the DxOmark has the expected behaviou, dropping by about a stop for each stop of ISO, the techradar is all over the place, and like no DR curve you ever saw. Conclusion, they are doing something very wrong, all their results are valueless.

My guess is there is uncontrolled processing of some sort in the generation of the TIFF.

--
Bob

ihv · Apr 6, 2012

My general feeling is that Canon is somewhat struggeling.

While there are for sure a lot of happy users with the 5D3, it also has created a lot of mixed feelings. I'm sure the product will do financially fine, but the cost is in brand - just adding the "Canon" label will not create as much sales in the future as it used to be, I think.

kikiriki said:
Can they do high MP sensor? Yes, they did it..

http://www.dpreview.com/news/2010/8/24/canon120mpsensor

It is just matter of timing, and some people will never understand that..

In business timing is everything, and Canon team for some reason think it is better to see what competition can do, and then answer with clear targets..
--
Igor Leskovar

RedFox88 · Apr 6, 2012

Silverback1988 said:
Hi carlk,
I was a little disappointed with 5DIII,
I thought that camera was my upgrade path for my 5DII.
I am waiting for a 48 MP (or so) from Canon.
And I believe they could deliver that,
because their current technology is up to it.
Just look their 18MP APSC cameras,
IF their sensor area is as big as 35mm FF,
THEN The IQ would surpass the 5DIII.

Why are you confusing resolution for image quality? They are separate things!

Silverback1988 said:
I don't think a single exposure from 5DII or 5DIII would match 2 exposures from 7D (stitched) at ISO 100.

So.. IMHO... Canon could make one that better than 5DIII (from IQ stand point).
The question is, would they make one ?
First time, I was doubt it,
but with the release of 60Da that designed for a specific task,

I consider it as a good sign from Canon that they would make a specific camera for another task : landscape / fashion / studio which need higher resolution.

Let's see when that happen.

Now, I still using 5DII which still a great tools for me, not interest with 5DIII for an upgrade.
-
Brian

carlk said:

Canon apologist would say the 5DIII sensor is not a bad one and an improvement over the previous gen but I don’t see it that way. Yes it’s not a bad one if the world has been standing still but the technology landscape is very different now. One more MP and perhaps slightly better noise with no improvement in DR for three and half years while the rest of the world has been progressing in leaps and bounds just put them in such a bad position.

Three and half years ago Canon’s 21MP sensor had absolutely no peer. Few short years later everyone from phone camera to dslr sensor makers are able to develop much better sensors except for Canon. iPhone at that time had a noisy measly 2MP sensor. It now has an 8MP sensor that is able to take very nice pictures even in relatively low lights. Nokia’s 41MP sensor with varies crop and down sampling modes produces great low noise still and full HD video images. It even beats the 5DIII sensor in video over-sampling amount. Not to be lost here is those are sensors come with phones which cost just a small fraction of the high end dslr and can also make phone calls or surf the web.

On the higher end front Sony have the nice 36MP full frame sensor they made for Nikon camera and perhaps their own later. Even their aps-c sensor has more MP than Canon’s full frame sensor with IQ not far behind, or even exceeds it in certain areas. Fuji’s innovative aps-c sensor in X Pro 1 was said to be as good or better than most full frame dslr sensor for high or low ISO IQ. Imagine what if Canon have technologies similar to what Sony, Fuji and those phone sensor makers have to put into the latest camera? Is Canon supposed to be the leader or we are expecting too much?

Click to expand...

rwbaron · Apr 6, 2012

I see your point but it appears the 5D3 and 1Ds3 curves follow logic. If those curves are representative and they're being done by the same people in the same way with the same software why would the curves for the other cameras be off?

It's my understanding that DxoMark measures the RAW file data but in this case we're looking at the results from RAW's converted to Tiff's. On the surface it appears the 5D3 and 1Ds3 Tiff curves correspond to their RAW curves but not the case for the other cameras.

edit;

Do you believe the same holds true for the SNR values?

Bob

bobn2 said:
rwbaron said:

Interesting results using DXO's software and procedure to measure DR and SNR.

Now we need the debate as to why this doesn't correspond with the 2+ stops of DR the D800 is supposed to have over the 5D3. At ISO 100 to about 700 the Tiff's are showing less than a 1/2 stop advantage for the D800 and past 800 the 5D3 is better by about the same value. Furthermore, the chart shows the Tiff SNR of the 5D3 to be higher than the D800 across all ISO's except for 800. Are these people that inept or is there something the techies on this forum have missed?

I'm very interested in the explanation.

Click to expand...

Those are very strange curves. Compare their curve

with DxO's DR curve for the D800
100: TR has 12.7, DxO 13.23
200: 12.6, 12.56
400, 12.6, 11.74
800, 11.7, 10.93
1600, 11.7, 10.03
3200, 10.2, 9.0
6400, 9.3, 8.07
12800, 8.7, 7.06

So, while the DxOmark has the expected behaviou, dropping by about a stop for each stop of ISO, the techradar is all over the place, and like no DR curve you ever saw. Conclusion, they are doing something very wrong, all their results are valueless.

My guess is there is uncontrolled processing of some sort in the generation of the TIFF.

--
Bob

--
http://www.pbase.com/rwbaron

altair8800 · Apr 6, 2012

kikiriki said:
Can they do high MP sensor? Yes, they did it..

http://www.dpreview.com/news/2010/8/24/canon120mpsensor

It is just matter of timing, and some people will never understand that..

In business timing is everything, and Canon team for some reason think it is better to see what competition can do, and then answer with clear targets..
--
Igor Leskovar

So, they decided that leading as they did in the past was bad for their image? Announce 120Mp prototypes and stick with low res FF products? Yes, some people will never understand their timing, including me.

Dan

bobn2 · Apr 6, 2012

rwbaron said:
I see your point but it appears the 5D3 and 1Ds3 curves follow logic. If those curves are representative and they're being done by the same people in the same way with the same software why would the curves for the other cameras be off?

I don't know. I can only think it is in the TIFF. Looking through their reviews many of them show these weird shape curves. I suspect their competence.

rwbaron said:
It's my understanding that DxoMark measures the RAW file data but in this case we're looking at the results from RAW's converted to Tiff's. On the surface it appears the 5D3 and 1Ds3 Tiff curves correspond to their RAW curves but not the case for the other cameras.

Well, we don't know, do we, since we haven't seen any reliable data for them. Having looked through their reviews now, they are in my 'pinch of salt' list.
--
Bob

Horshack · Apr 6, 2012

rwbaron said:
http://www.techradar.com/news/photography-video-capture/cameras/noise-and-dynamic-range-results-explained-1027588

Where's the difference?

How the images were processed, the ongoing calibration of the light source, experience.

What happened to Canon sensor development?

Forum Pro

Veteran Member

Forum Pro

Senior Member

Forum Pro

Forum Pro

Forum Pro

Forum Pro

Active member

Forum Pro

Well-known member

Senior Member

Forum Enthusiast

Forum Pro

Leading Member

Forum Pro

Forum Pro

Senior Member

Forum Pro

Forum Pro

Similar threads

Keyboard shortcuts