WOW, 5D III shadow recovery!!

[jrista]

I was browsing through some of the 5D III sample shots from CPN Europe. I came across this one, which appears to be an in-camera multi-exposure JPEG.

http://cpn.canon-europe.com/files/education/technical/inside_canon_eos_5d_mark_iii/17_cinc_big.jpg

I was curious if there was any fixed-pattern read noise (banding noise) in the silhouetted parts of the image, so I downloaded it and started adjusting levels in Photoshop. The image below is a composite of two strips that have mid-gray levels adjusted by 3.4, next to two strips of the original image. I am blown away at the black recovery the 5D III has to offer!!! I mean, WOW:

 

[jrista]

Wow, seriously...no one has any comments?

 

[Daniel Flather]

Wow, seriously...no one has any comments?

I looks like it's epic omg lol zomgers!

 

[jrista]

Wow, seriously...no one has any comments?

I looks like it's epic omg lol zomgers!

Heh...honestly, I was expecting something more along the lines of: "D3s/D700 is still better. Check this image out <yaddayaddaimglink>".

 

[Daniel Flather]

Wow, seriously...no one has any comments?

I looks like it's epic omg lol zomgers!

Heh...honestly, I was expecting something more along the lines of: "D3s/D700 is still better. Check this image out <yaddayaddaimglink>".

No, no, you want the omg it's too expensive but I never was gonna buy it anyways!! omg canon failz0r

 

[Tuggem]

I'm impressed by what I see.
Just not sure of what conclusions to make if it was a multi exposure to start with?

 

[tt]

Looking forward to multi exposure. - the composite looks impressive.
I kinda like the concept of the photo multi exposure you've made to in a stylistic way

 

[jrista]

I'm impressed by what I see.
Just not sure of what conclusions to make if it was a multi exposure to start with?

Hmm, thats actually a good point...stacking shots and blending them could lower noise.

 

[jrista]

Looking forward to multi exposure. - the composite looks impressive.
I kinda like the concept of the photo multi exposure you've made to in a stylistic way

Oh, its not mine. Its just one of the sample shots from Canon's site, I just tweaked the exposure a bit while I was looking for banding noise, and discovered there was a LOT of color detail in all the original black.

 

[Kernuak]

I did glance at that one earlier in my quest to look for the high ISO samples, then didn't look any harder at it when it was "only" ISO 400. It's difficult to be sure in JPEGs, but I'm leaning towards a two stop improvement. I'd even be interested in seeing some ISO 25600 shots without the heavy in camera noise reduction, as I think they could look alot better than the sample does. It is certainly impressive, but not entirely a surprise to me, as I was think about the 7D, the native was basically two stops higher than I'd like to push it, so working on the same principal, I was thinking the MkIII might be two stops better than the MkII. I also saw much more random noise in the 7D than in previous cameras and less banding.

 

[jrista]

I did glance at that one earlier in my quest to look for the high ISO samples, then didn't look any harder at it when it was "only" ISO 400. It's difficult to be sure in JPEGs, but I'm leaning towards a two stop improvement. I'd even be interested in seeing some ISO 25600 shots without the heavy in camera noise reduction, as I think they could look alot better than the sample does. It is certainly impressive, but not entirely a surprise to me, as I was think about the 7D, the native was basically two stops higher than I'd like to push it, so working on the same principal, I was thinking the MkIII might be two stops better than the MkII. I also saw much more random noise in the 7D than in previous cameras and less banding.

Aye, I really want to see some RAW samples myself...JPEG kind of muddies the picture a bit (sorry for the pun.) We don't really know how much noise reduction and other fancywork is being done in-camera for these sample JPEGS.

The 7D still has banding noise, its just not as strong as on the 5D II (the 7D generates about 9 electrons worth of read noise while the 5D II generates about 27 electrons.) Based on the jpegs, I'd say the 5D II (and probably the 1D X) produce as little read noise as Sony Exmor sensors now (which is about 3 electrons worth)...I've cranked the exposure up on every dark and high ISO shot from all the 5D III samples I could find, and there isn't a trace of banding noise in any of them. Cranking up exposure on my 7D, it doesn't take long to start seeing banding in shadowed areas, and it really destroys detail where SNR is lower.

One thing I have noticed in the 5D III shots, though, is quite a bit of color blotchieness in low SNR areas. This was a problem before on Canon sensors, and it still seems to be a problem, although its not bad enough to really affect most work until you hit about ISO 6400...which is definitely an improvement. On my 7D, it starts to intrude at about ISO 1600, and is really bad by ISO 6400. Sony Exmor sensors still seem to do a better job on this front (D3s 12800 photos exhibit random noise but very, very little color noise.)

 

[Mt Spokane Photography]

I downloaded the original jpeg files from all the ISO settings on DPR, and tried pulling up the blacks. Even with the slider all the way, there was no banding, I was quite amazed. Canon has really worked on the blacks and the ability to pull out detail.

 

[Curmudgeon]

It was making me a little nervous that CW and other Canon spokesmen spent a lot of time talking about at least two stops of high ISO improvement, but none of them was actually addressing the matter of dynamic range, which for me is of greater interest than simple high-ISO capabilities. Likewise, the new HDR feature, while admittedly very intriguing in its own right, had me wondering if maybe Canon had given up trying to match the DR of Sony's Exmor sensors. We really do need some definitive testing on IQ, but as little bits and pieces emerge about the 5D3's performance, it's looking more encouraging all the time. Maybe I really do need one now rather than later.

Oh, and jrista, you have a history of introducing slow-starting threads.

 

[LetTheRightLensIn]

You can't really tell until you see the RAWs.

Although at high iso it seems like there is a decent chance they did get rid of the banding.

 

[Mt Spokane Photography]

You can't really tell until you see the RAWs.

Although at high iso it seems like there is a decent chance they did get rid of the banding.

Yes, I'm waiting for them. I am a bit skeptical, but hopeful.

However, I would expect to extract better performance out of a raw, and the jpegs already surpass my 5D MK II raws by quite a jump.

 

[Arjay]

Just to add to this, on the still life with berries, I tried raising the shadows to a ridiculous amount.
Not bad at all.

 

[dafrank]

Just to add to this, on the still life with berries, I tried raising the shadows to a ridiculous amount.
Not bad at all.

Just so happened I did the same thing with the same downloaded image yesterday, and I too was very impressed at the lack of noise and banding in areas of shadow that were lifted at least two f/stops. All the more impressive was that the image was saved as a 8-bit Png, a pretty lousy basis for heavy-duty file manipulation. I'm looking forward to seeing more from the camera. Unfortunately, Canon's sample shooters on the 5DIII have done a miserable job so far. If I can verify the performance that I suspect, but have not yet seen with my own eyes, I'll be buying a couple of them.

 

[LetTheRightLensIn]

The thing about low ISO jpgs is that if they buried the banding and muck then no matter how you raise the jpg it won't show even if it might be there in spades in the RAW so I don't think we can say for sure yet. However, there appears to be promise for high iso banding, although it's again hard to say for sure yet.

One thing that worries me is that every single mention about improved DR has specifically stated at mid to high iso settings. Does that imply it still has the same nasty non-linear DR plot where you don't gain much DR below ISO400 or 800 and it will fall 2+ stops shy of the SOny Exmor sensors used in the latest Nikons again?? I hope not but, but if they had fixed that wouldn't they be expounding on it? Instead they seem to be specifically talking about high iso improvements only.

I have a weird feeling the 36Mp D800 might end up with 2-3 stops better DR at ISO 100 than the 22MP 5D3. I really hope I am wrong though.

 

[jrista]

I have a weird feeling the 36Mp D800 might end up with 2-3 stops better DR at ISO 100 than the 22MP 5D3. I really hope I am wrong though.

What would make you suspect that? At best, you can get 14 stops out of a 14bit sensor. The previous generation of Canon sensors reached 11.7 stops at best, and around 11.2 on average...despite having 14bit sensors. Sony Exmor sensors in Nikon cameras get about 13.8 stops. If the 5D III has about two stops better high ISO, then you have to figure DR has improved to a degree. Based on high ISO samples from the 5D III, there does not appear to be any banding noise at all, and that was what was eating away the remaining two stops or so of DR from the 5D II (thanks to a noisy ADC, something Canon's press releases and information on their site claims they have specifically addressed). I would be willing to bet the 5D III gets at least 12.8 stops, if not somewhere in the same vicinity as a Sony Exmor sensor.