New Canon 5D mark III raws

[ippikiokami]

Played with the 25600 raw a little bit in Topaz Denoise and all I can say is Sweet Baby Jesus this generation of cameras is amazing.

If the AF is as accurate as it should be it's a major upgrade from my 5d II (which i'm happily keeping btw. It didn't all the sudden start taking crap pictures). I was going to bite the bullet and grab a 1DX but I might skip a 1 series generation this time around

 

[jrista]

(The D800 would offer 50% more MP and perhaps a solid 2.5 stops better dynamic range at ISO 100 and perhaps between 1/3 stop worse to 1 stop better performance at high ISO. Based on that, the D800 most likely does have an all around better sensor, although it is not set in stone yet.*)

How do you figure the D800's DR is 2.5 stops better than the 5DIII? Just curious.

He can't. Theoretically, the maximum possible DR would be 14 stops with a full 14-bit image sensor and image processing pipeline (think about the nature of a bit...every successive bit has twice the significance...or value-holding power...as the previous; a doubling; in other words, every bit is one EV, or one stop, difference in dynamic range.) The 5D II achieved about 11.86 stops of DR, and the D7000 (Nikon's highest DR camera) achieved about 13.87 stops of DR (based on DXO data, which measures DR from the point where a single photon strikes the sensor to the point where the first pixel is fully saturated). Assuming the 5D III has not improved at all on the DR front, the maximum difference in DR would be about 2.01 stops. I think most of us are pretty solidly confident that Canon has resolved their read noise issues, and are probably getting much closer to that 13.9 stops of maximum DR that Sony Exmor sensors are getting...so the difference is probably less than a stop, (personally I hope and believe it will be in the realm of 0.25 or less stops), of DR difference between any one of the 1D X, 5D III, D800, D4 and D7000.

 

[LetTheRightLensIn]

(The D800 would offer 50% more MP and perhaps a solid 2.5 stops better dynamic range at ISO 100 and perhaps between 1/3 stop worse to 1 stop better performance at high ISO. Based on that, the D800 most likely does have an all around better sensor, although it is not set in stone yet.*)

How do you figure the D800's DR is 2.5 stops better than the 5DIII? Just curious.

Because someone measured the ISO100 read noise of the 5D3 (using masked area of the image, granted that is not ideal) and measured essentially the same exact read noise as for the 5D2 and because the D800 uses and Exmor sensor, most of which have had a good 2-2.5 stops on the 5D2 (and the difference is even greater since they are also free of pattern banding which bothers the eye more).

We will see, but things, unfortunately, seem to be pointing this way. I hope it won't be the case, but....

 

[LetTheRightLensIn]

I downloaded the raws but my photoshop cs5 cant load them nor DPP.. what update do I need to do this? Regarding image resource, I compared apples to apples Canons 5d3 vs Canons 5d2 on same images and the 5d3 looked sharper overall in almost all the ISO ranges compared to the 5d2, which seems to differ from internet sample images. All things being the same, if this is indeed the case that the 5d3 is cleaner and sharper than the 5d2, i'm game. It's a shame they dont have a D800 to compare against so we can really comb down to brass tax and know where we stand...

If you search around on google you can find the link for ACR 6.7RC.

Yeah the comparator on IR gives a very false impression since you are comparing 5D2 and 5D3 images that were each processed with extremely different settings and different converter version, the huge sharpness advantage goes away look a images processed with the same settings.

 

[SRHelicity]

Same could be said about Canon as well. Who's to say that Canon won't actually release the rumored 5DX. High 30s - low 40s MPs.

My problem is that, if the 22 MP 5D3 is $3500, there's no way a high MP 5DX would be anything less than $4000. Sure, maybe they'll use a less sophisticated AF and metering system, but I can't find any way for Canon to compete with the D800 in terms of price if the 5D X comes to pass. I mean, do you think Canon would price a 30-40 MP FF camera BELOW the price of the 5D Mk III? I don't.

 

[LetTheRightLensIn]

(The D800 would offer 50% more MP and perhaps a solid 2.5 stops better dynamic range at ISO 100 and perhaps between 1/3 stop worse to 1 stop better performance at high ISO. Based on that, the D800 most likely does have an all around better sensor, although it is not set in stone yet.*)

How do you figure the D800's DR is 2.5 stops better than the 5DIII? Just curious.

He can't. Theoretically, the maximum possible DR would be 14 stops with a full 14-bit image sensor and image processing pipeline (think about the nature of a bit...every successive bit has twice the significance...or value-holding power...as the previous; a doubling; in other words, every bit is one EV, or one stop, difference in dynamic range.) The 5D II achieved about 11.86 stops of DR, and the D7000 (Nikon's highest DR camera) achieved about 13.87 stops of DR (based on DXO data, which measures DR from the point where a single photon strikes the sensor to the point where the first pixel is fully saturated). Assuming the 5D III has not improved at all on the DR front, the maximum difference in DR would be about 2.01 stops. I think most of us are pretty solidly confident that Canon has resolved their read noise issues, and are probably getting much closer to that 13.9 stops of maximum DR that Sony Exmor sensors are getting...so the difference is probably less than a stop, (personally I hope and believe it will be in the realm of 0.25 or less stops), of DR difference between any one of the 1D X, 5D III, D800, D4 and D7000.

Oops I used my non-normalized 11.25 value for the 5D2 and mixed it with the 8MP normalized value for the D7000. So yeah with the 11.8 then you are at 2 stops difference. But don't forget that the D7000 is superbly free of any fixed pattern noise while the 5D2 was among the worst of Canon's efforts in that regard so the usable difference has got to be at least another 1/2 stop more. So you still talking a good 2.5 stops better in a real-world usable sense.

Perhaps the D800 won't quite match the D7000 and perhaps the main sensor area will measure a bit better for the 5D3 and it will come out closer than 2 stops. I hope the 5D3 does better than the first result hinted at because it was kind of disappointing and not what I had hoped for.

 

[Orion]

I am at a loss for words when comparing the 5DmkIII to the D4. . . . @ 6400iso and 12800 iso I viewd full res at the base of the goblin beer bottle, and the 5DmkIII kicks arse. I find it hard to believe though. . . . I wasn;t expecting so much noise from the D4 and sharper image from teh mkIII just from that section of the image. . . .

 

[LetTheRightLensIn]

I think most of us are pretty solidly confident that Canon has resolved their read noise issues, and are probably getting much closer to that 13.9 stops of maximum DR that Sony Exmor sensors are getting...so the difference is probably less than a stop, (personally I hope and believe it will be in the realm of 0.25 or less stops), of DR difference between any one of the 1D X, 5D III, D800, D4 and D7000.

I had hoped that but earlier today someone on DPR measured 6.02 ADU for the 5D3 at ISO 100 while my 5D2 measures 6.09 ADU. So nothing there. Basically the exact same. I just hope the 5D3 is one of the few bodies where the masked area performs much differently from the main area. But it seems more like it will be 11.8 going to 12.05 versus 11.8 going to 13.5 or 13.8 and with the 5D3 having more fixed pattern noise the usable differences ending up over 2 stops worse. I haven't measured it yet myself, I just hope the other guy messed it up (he is usually not one to miss this stuff up though) or that the masked area just doesn't work well for measuring this body. If we are really lucky ISO 100 will have been rated 1/2 stop off compared to the 5D2 and then we can gain an extra 1/2 stop back and 12.05 goes to 12.55 but that is probably not going to be the case. Anyway it seems unlikely the 5D3 will be able to measure better than 12.6 and it might be 11.9 again. The D800 would probably be 13.5-13.9 plus have less fixed pattern noise, so whatever it has will have an extra bonus of further usability.

 

[3kramd5]

My problem is that, if the 22 MP 5D3 is $3500, there's no way a high MP 5DX would be anything less than $4000. Sure, maybe they'll use a less sophisticated AF and metering system, but I can't find any way for Canon to compete with the D800 in terms of price if the 5D X comes to pass. I mean, do you think Canon would price a 30-40 MP FF camera BELOW the price of the 5D Mk III? I don't.

Maybe, maybe not. Depends at least partially on development and production costs. If they scale an existing sensor (like the 7D's) to 36X24 (45MP-ish) and use less expensive AF and processing, it's possible they could produce a high pixel count camera for less than the 5D3 (entirely new sensor I believe).

 

[mrmarks]

Is there any noise banding in the 5D3?

 

[jrista]

I think most of us are pretty solidly confident that Canon has resolved their read noise issues, and are probably getting much closer to that 13.9 stops of maximum DR that Sony Exmor sensors are getting...so the difference is probably less than a stop, (personally I hope and believe it will be in the realm of 0.25 or less stops), of DR difference between any one of the 1D X, 5D III, D800, D4 and D7000.

I had hoped that but earlier today someone on DPR measured 6.02 ADU for the 5D3 at ISO 100 while my 5D2 measures 6.09 ADU.

Who did that measurement? If your talking someone on the forums, thats not what I'm referring to. If it was an official step wedge test based on JPEG output (which are usually what official DPR DR tests are based on), those tests are generally worthless (JPEG compression obliterates DR.) The DXO tests are very accurate, and measure the sensor hardware at a lower level than a JPEG step wedge test does. From a headroom standpoint, i.e. the ability to recover highlights or shadows, DXO's tests are more accurate (as demonstrated by the many videos on the net showing unbelievable shadow recovery with the D7000, where you can see before your eyes the recovery of 4-6 stops of what appears to be solid black in under exposed photos.)

So nothing there. Basically the exact same. I just hope the 5D3 is one of the few bodies where the masked area performs much differently from the main area. But it seems more like it will be 11.8 going to 12.05 versus 11.8 going to 13.5 or 13.8 and with the 5D3 having more fixed pattern noise the usable differences ending up over 2 stops worse. I haven't measured it yet myself, I just hope the other guy messed it up (he is usually not one to miss this stuff up though) or that the masked area just doesn't work well for measuring this body. If we are really lucky ISO 100 will have been rated 1/2 stop off compared to the 5D2 and then we can gain an extra 1/2 stop back and 12.05 goes to 12.55 but that is probably not going to be the case. Anyway it seems unlikely the 5D3 will be able to measure better than 12.6 and it might be 11.9 again. The D800 would probably be 13.5-13.9 plus have less fixed pattern noise, so whatever it has will have an extra bonus of further usability.

Where in the world are you getting that the 5D III has MORE fixed pattern noise? So far, in all my poking around with the samples available on the net, the 5D III has no visible fixed pattern noise at all as far as I can tell, right down into the deep shadows. You got a link somewhere that shows the 5D III having WORSE fixed pattern noise than the 5D II (which was pretty bad even for its time, at 27.8 electrons worth.)

 

[justsomedude]

My problem is that, if the 22 MP 5D3 is $3500, there's no way a high MP 5DX would be anything less than $4000. Sure, maybe they'll use a less sophisticated AF and metering system, but I can't find any way for Canon to compete with the D800 in terms of price if the 5D X comes to pass. I mean, do you think Canon would price a 30-40 MP FF camera BELOW the price of the 5D Mk III? I don't.

It really depends on the market they're going after. Remember, the 5D3 is 22MP, with 6fps and an insane AF system with killer ISOs at $3499. A 5D body with lower fps, say 3-4 frames per second, a 30+ MP sensor, less advanced AF system and ISOs comparable to the d800 could certainly be priced in the $2799-$2999 range to compete directly with Nikon's 36MP beast.

Remember, it's not just megapixels a body is packing, but other features as well that photographers look for. High MP shooters don't always need high burst and high ISOs, so perhaps the 5D3 is not for them. A 5DX could satisfy that market at $2,899 while leaving the 5D2 intact at $2,199.

Time will tell.

 

[LetTheRightLensIn]

Is there any noise banding in the 5D3?

It seems like the horizontal banding is mostly gone low and high iso but vertical is still here. It doesn't appear to show up much at high iso (although we don't have any really dark samples yet) but is still in low ISO shadows to some degree or another, it seems to be more than the 1Ds3 has.

 

[LetTheRightLensIn]

I think most of us are pretty solidly confident that Canon has resolved their read noise issues, and are probably getting much closer to that 13.9 stops of maximum DR that Sony Exmor sensors are getting...so the difference is probably less than a stop, (personally I hope and believe it will be in the realm of 0.25 or less stops), of DR difference between any one of the 1D X, 5D III, D800, D4 and D7000.

I had hoped that but earlier today someone on DPR measured 6.02 ADU for the 5D3 at ISO 100 while my 5D2 measures 6.09 ADU.

Who did that measurement? If your talking someone on the forums, thats not what I'm referring to. If it was an official step wedge test based on JPEG output (which are usually what official DPR DR tests are based on), those tests are generally worthless (JPEG compression obliterates DR.) The DXO tests are very accurate, and measure the sensor hardware at a lower level than a JPEG step wedge test does. From a headroom standpoint, i.e. the ability to recover highlights or shadows, DXO's tests are more accurate (as demonstrated by the many videos on the net showing unbelievable shadow recovery with the D7000, where you can see before your eyes the recovery of 4-6 stops of what appears to be solid black in under exposed photos.)

Not a wedge test. Measuring the masked off black area of an ISO 100 RAW, finding the std dev in raw levels, then taking the max raw level - the raw black point and then dividing it by the read noise and then taking the log2 of that to get dynamic range in stops. JS provided the raw read of 6.02 ADU. It's the same thing DxO does and same thing I did in my tests of the 40D,20D,50D,7D,5D2,1D3.

 

[LetTheRightLensIn]

Where in the world are you getting that the 5D III has MORE fixed pattern noise? So far, in all my poking around with the samples available on the net, the 5D III has no visible fixed pattern noise at all as far as I can tell, right down into the deep shadows. You got a link somewhere that shows the 5D III having WORSE fixed pattern noise than the 5D II (which was pretty bad even for its time, at 27.8 electrons worth.)

No, not worse than the 5D2 but worse, at low ISO, than the 1Ds3 and the D7000, D3x, almost certainly the D800, etc. It seems to be a little bit better than the 5D2 but it appears to still be there worse than any of Canon's best efforts, most likely, and almost 100% sure worse than any of Sony/Nikon's best efforts.

It seems like it won't be a problem at high iso though.

 

[LetTheRightLensIn]

I think most of us are pretty solidly confident that Canon has resolved their read noise issues, and are probably getting much closer to that 13.9 stops of maximum DR that Sony Exmor sensors are getting...so the difference is probably less than a stop, (personally I hope and believe it will be in the realm of 0.25 or less stops), of DR difference between any one of the 1D X, 5D III, D800, D4 and D7000.

I had hoped that but earlier today someone on DPR measured 6.02 ADU for the 5D3 at ISO 100 while my 5D2 measures 6.09 ADU.

Who did that measurement? If your talking someone on the forums, thats not what I'm referring to. If it was an official step wedge test based on JPEG output (which are usually what official DPR DR tests are based on), those tests are generally worthless (JPEG compression obliterates DR.) The DXO tests are very accurate, and measure the sensor hardware at a lower level than a JPEG step wedge test does. From a headroom standpoint, i.e. the ability to recover highlights or shadows, DXO's tests are more accurate (as demonstrated by the many videos on the net showing unbelievable shadow recovery with the D7000, where you can see before your eyes the recovery of 4-6 stops of what appears to be solid black in under exposed photos.)

TheLensRightIn appears to be a Nikon troll. Just smite him and move on, ignoring further posts. Stop feeding the troll.

I'm ignoring anyone making claims of DR being anything until I see results from DxO.

Dude, I've owned Canon SLRs since the EOS 650.... just saying. But be my guest and spend your every waking 4 hour period re-smiting me, if it makes you feel better, I couldn't care less.

If someone can provide a 1/8000the exposure with the body cap on in a dark room and a frame with all the channels totally blown out, both at ISO 100 then we can get a better measurement. This DR measurement is easy. When it I do it with my 5D2 I get virtually the same result DxO reports.

 

[LetTheRightLensIn]

Well anything up to and including ISO 3200 appears to be relatively free of noise.

At least to my eyes, ISO 3200 on the 5D Mark III is about equivalent to ISO 400 on the 5D Mark II, so a 3 stop improvement based on what I'm seeing.

That seems a bit much but anything over ISO 400 on the 5D Mark II starts to show noise and the 5D Mark III controls it well until ISO 6400.

So I'm saying there's at least two stops improvement in noise from the 5D Mark II to the 5D Mark III and perhaps three.

5D Mark II ISO 400:
http://www.imaging-resource.com/PRODS/E5D2/FULLRES/E5D2hSLI00400_NR_OFF.HTM
5D Mark III ISO 3200:
http://www.imaging-resource.com/PRODS/canon-5d-mkiii/FULLRES/E5D3hSLI003200.HTM

Hmmm, it just occurs to me that I might be comparing "noise reduction off" images with "noise reduction on" images. I can't see anything in the 5D Mark III image's EXIF to indicate if or what noise reduction was set to. Anyone?

Yes, you are comparing images that were rendering with a different converter using entirely different settings.
You need to download the RAW files and look at them in Photoshop yourself using the same settings.

 

[jrista]

I think most of us are pretty solidly confident that Canon has resolved their read noise issues, and are probably getting much closer to that 13.9 stops of maximum DR that Sony Exmor sensors are getting...so the difference is probably less than a stop, (personally I hope and believe it will be in the realm of 0.25 or less stops), of DR difference between any one of the 1D X, 5D III, D800, D4 and D7000.

I had hoped that but earlier today someone on DPR measured 6.02 ADU for the 5D3 at ISO 100 while my 5D2 measures 6.09 ADU.

Who did that measurement? If your talking someone on the forums, thats not what I'm referring to. If it was an official step wedge test based on JPEG output (which are usually what official DPR DR tests are based on), those tests are generally worthless (JPEG compression obliterates DR.) The DXO tests are very accurate, and measure the sensor hardware at a lower level than a JPEG step wedge test does. From a headroom standpoint, i.e. the ability to recover highlights or shadows, DXO's tests are more accurate (as demonstrated by the many videos on the net showing unbelievable shadow recovery with the D7000, where you can see before your eyes the recovery of 4-6 stops of what appears to be solid black in under exposed photos.)

Not a wedge test. Measuring the masked off black area of an ISO 100 RAW, finding the std dev in raw levels, then taking the max raw level - the raw black point and then dividing it by the read noise and then taking the log2 of that to get dynamic range in stops. JS provided the raw read of 6.02 ADU. It's the same thing DxO does and same thing I did in my tests of the 40D,20D,50D,7D,5D2,1D3.

To date, from ISO 100 to ISO 25600 test images, both JPEG and RAW, not one SOUL outside of you has mentioned any amount of banding noise visible in ANY image from the 5D III (most of which are from pre-production models even!) From the noise levels at ISO 6400, even at 25600, the 5D III, with black and shadow levels cranked up as high as they can go in Photoshop, the only thing visible even in the darkest blacks is clean, random noise. I cranked up EV on every ISO 100 image I could get my hands on, and I have seen nothing but a few speckles of random noise in the darkest shadows. You would need to produce an actual sample image that clearly demonstrates banding noise before I'd even consider that you are not purely fabricating the details our producing (i.e. 6.02 ADU.) You would also need to produce a link to whatever forum thread or article on DPR that you are referring to that supposedly measured the 5D III's read noise, because I'm highly skeptical anyone has had a chance to do any accurate measurements with the proper gear and care to produce anything remotely resembling a valid result.

Additionally, if you are trying to replicate DXO tests on your own with Photoshop/ACR, Lightroom, Aperture, or any other mainstream RAW editor, your not working with accurate data. Outside of DCRAW, every RAW processor applies a tone curve (picture style/image style/etc.) to the RAW image data before rendering to screen or file. You aren't looking at the unmodified RAW image data, even if you use a "neutral" tone curve...since even neutral is non-linear. You need to use the right kind of image processing software, as well as specially designed test targets, to test the full DR capable by a digital sensor. DXO not only uses a very specialized system to measure noise and DR (http://www.dxomark.com/index.php/About/In-depth-measurements/DxOMark-testing-protocols/Noise-dynamic-range), they use the same system to measure every camera, so comparisons are accurate.

 

[justsomedude]

Not a wedge test. Measuring the masked off black area of an ISO 100 RAW, finding the std dev in raw levels, then taking the max raw level - the raw black point and then dividing it by the read noise and then taking the log2 of that to get dynamic range in stops. JS provided the raw read of 6.02 ADU. It's the same thing DxO does and same thing I did in my tests of the 40D,20D,50D,7D,5D2,1D3.

Just curious - how are you doing these measurements? ACR and LR don't even support the 5D3 RAW files yet... at least on my end I'm stopped by camera compatibility warnings when I try to import into either program.

 

[jrista]

@justsomedude: You can try this: http://labs.adobe.com/downloads/cameraraw6-7.html