Noise and DR comparision: Canon 5D vs 5D2 vs 7D vs Pentax K5

[Guillermo Luijk]

The above may not be 100% correct though. The Full SNR curves at DxO are pixel level curves (aka "screen") and have not been normalized to 8MP. So if I'm not mistaken, the DxO figure of 11.0 stops would compare to Guillermo's figure of 11.2 stops which still is a nice match.

I didn't explain well. My 11.2EV figure is DR after normalising to 5D resolution (i.e. 13Mpx), so can't be compared to any DxO figure.
My per-pixel based DR figure, obtained from these per-pixel plots:

Is a DR of 10.8EV for the Pentax, 0.2EV below the calculation from DxO curves which I found to be 11.0EV.

Regards

[Guillermo Luijk]

A comparision ISO100 vs ISO1600 (same aperture/shutter) in a high dynamic range scene with the Pentax K5:

Camera's JPEG at ISO100 (to have an idea of the scene's DR):


HDR tone mapped scene:


Crops in highlights, medium shadows and deep shadows:

• In the highlights the 4 extra stops allow to properly capture the lamp area and view through the window at ISO100, not at ISO1600
• In the medium shadows, no SNR improvement for pushing ISO
• In the deep shadows, it seems read noise becomes important and a bit more noise can be seen in the ISO100 image. Also many pixels get clipped to 0 in the RAW file (perhaps the 14-bit RAW encoding was not enough here) producing darker areas lacking detail that displays better in the ISO1600 shot

If I had to choose, I would pick the ISO100. But we can see there are still some advantages in the very deep shadows for pushing ISO.

Regards

PS: RAW files to play
ISO100: http://dl.dropbox.com/u/136377/ISO100.DNG
ISO1600: http://dl.dropbox.com/u/136377/ISO1600.DNG

[Ray]

A comparision ISO100 vs ISO1600 (same aperture/shutter) in a high dynamic range scene with the Pentax K5:

• In the highlights the 4 extra stops allow to properly capture the lamp area and view through the window at ISO100, not at ISO1600
• In the medium shadows, no SNR improvement for pushing ISO
• In the deep shadows, it seems read noise becomes important and a bit more noise can be seen in the ISO100 image. Also many pixels get clipped to 0 in the RAW file (perhaps the 14-bit RAW encoding was not enough here) producing darker areas lacking detail that displays better in the ISO1600 shot

If I had to choose, I would pick the ISO100. But we can see there are still some advantages in the very deep shadows for pushing ISO.

Regards

PS: RAW files to play
ISO100: http://dl.dropbox.com/u/136377/ISO100.DNG
ISO1600: http://dl.dropbox.com/u/136377/ISO1600.DNG

Excellent examples, Guillermo. You have saved me the trouble of testing this situation with my D7000.  ;D

The extra detail in the extreme highlights, in the pushed ISO 100 shot, seems far more dramatic than the very marginal, additional noise in the deepest shadows.

However, I just might do such a comparison with my D7000 to see if that possible additional noise in the deepest shadows obscures detail. But it doesn't seem a big deal at this stage.

[JR]

Thank you, Guillermo. Looks like an amazing sensor. I will definitely try the new K5 when our local dealer gets one.

- John

[feppe]

HDR tone mapped scene:

Is this tone mapped from a single ISO100 RAW file? If so, did you use special techniques (eg. processing one version for highlights, one for shadows, combining with Zero Noise), or just use curves in the RAW processor?

In any case, if this is from just 1 frame, wow. I hope they'll bring these sensors to MFT cameras soon.

[Guillermo Luijk]

Is this tone mapped from a single ISO100 RAW file? If so, did you use special techniques (eg. processing one version for highlights, one for shadows, combining with Zero Noise), or just use curves in the RAW processor?

In any case, if this is from just 1 frame, wow. I hope they'll bring these sensors to MFT cameras soon.

I think I fused the ISO100 and ISO1600 files with Zero Noise, then tonemapped the fusion with TuFuse (see some dark halos near the lamp, not rare on TuFuse). Even if I did that, tonemapping just the ISO100 capture would have produced pretty much the same result thanks to this sensor's high DR and the very limited improvement in noise when raising ISO.

[Guillermo Luijk]

I have done an experiment to try to find out if in such a low-noise sensor as the K5's, those 14 bits are really useful or just marketing like in most 14-bit cameras. I asked Manuel Llorens to compile a version of DCRAW allowing the user to discard the N least significant bits prior to demosaicing, and used it on the well known 6-stops underexposed DNG file from the K5:

It's difficult to see any loss from 14 to 13 bits, but using 12 bits visible posterization begins:

So the conclusion is that K5's sensor at ISO80 has such a low level of noise that can actually benefit from using 14-bit RAW files instead of staying at 12-bit, that would suffice most present 14-bit cameras.

Regards