A900 vs K5 - Landscape Photography

[deathvalleydave]

Hi David,

Ever since I saw that the Pentax K-5 got a DR score of 14.1, I've been wondering how I should keep this in perspective knowing my A900 "only" has a DR score of 12.3. I realize that you only get the 14.1 at ISO 80 on the Pentax, and then the figure starts to go down as you increase ISO. So, maybe my question should be: how do the K5 and A900 compare for DR from ISO 200 - 800? Or, because the K5 got a score of 14.1 at ISO 80, is this still a hugely significant step?

My dilemma right now is that I don't want to spend $1899 for the Zeiss 16-35, and you can get a K5 w/ Pentax 12-24 for almost the same cost... So, if you could shed any light on the DR of the K5, it would be greatly appreciated.

Thanks,
Dave

[pakodominguez]

My dilemma right now is that I don't want to spend $1899 for the Zeiss 16-35, and you can get a K5 w/ Pentax 12-24 for almost the same cost... So, if you could shed any light on the DR of the K5, it would be greatly appreciated.

Hi Dave,

Why don't you try the KM 17-35? at f:11 it is a really good lens and you can find it second hand for less than US$ 400.

Regards

Pako

[bakubo]

So, maybe my question should be: how do the K5 and A900 compare for DR from ISO 200 - 800?

You can compare the dynamic range of both cameras at various ISO values on the dxomark.com website.

[David Kilpatrick]

Theoretical DR is not of much use. I can tell you that no Canon ever made, 14-bit conversion of not, has the same capacity for highlight recovery in a correctly exposed image as the A900. That's because in photon/sensel terms, Canon is already overexposing and users even suggest that they should do more - which is why Canon introduced Highlight Tone Priority (a mode which effectively forces underexposure on the user).

Ultimately, the entire midrange of any digital file must be converted to a linear transform over a 6-bit span. That is what used to be the straight line portion of the sensitometric curve. If it is not converted to something around 6 to 7 bits maximum for this line, the image will look hopelessly flat and desaturated. The remaining bits of the whole range may be used for shadow detail, or highlight detail; it does not matter much where you position the straight line of the curve, but it does matter how you distribute the spare capacity.

Using 14 bits and a gain which places the exposure to the right, by definition, lowers the apparent noise level and improves many aspects of practical results but it greatly reduces the highlight headroom. The Pentax may in theory have one more stop in headroom, but I seriously doubt it as Pentax has never done down that route. Unless they equalised the RGB filter array densities and created a 'native uniWB for daylight' sensor, resulting in a camera with much lower apparen ISO ratings, their 14 bits will most likely all be used making the sensor achieve double the apparent high ISO of the Sony, which will mean less headroom at lower ISOs.

One of the good things about the A55's mirror is that it has allowed a true ISO 100 not a saturated ISO 100. Other aspects not so wonderful.

Since Canon has not been able to produce raw files with better real-world adjustment potential than A900 files, despite 14-bit, I will wait to see whether Pentax raw files do any better before assuming extra bit depth and DR really means a better file when adjusted for a good final image.

David

[agorabasta]

12-bit would be enough only if no serious transformations to the raw data were needed. In most cases it's almost so. But in reality the WB plus calibration adjustments eat 2-4 bits off the 12-bit precision. Add to the losses the non-linearities intentionally built into the systems, and you readily get posterized shadows at low ISO with a900 or you get dull highlight tones with a700.

The real answer is not the 14/16bit precision, it's rather the FP precision. The files may stay small enough that way, but we need more computing power. In fact, the files may stay smaller than the current 12-bit ones simply since there's no need for more than two FP digits in the deepest shadows and no need for more than 10 FP digits in the highlights. And then as sensels get smaller, capturing less photons per sensel, we also need progressively less precision in the highlights as the sensel full-well capacity diminishes with size. So at some point an extra gigapixel won't increase the file size any more
And only then we bump into the theoretical limits of the optics.

I really hope we'll get FP precision pretty soon.

[deathvalleydave]

David,

Regarding Canon's Highlight Tone Priority mode, does it only apply to their in camera jpegs?

If I understand you correctly, although Canon is using 14 bit RAW, they are also over-exposing the image which allows for less highlight recovery. Is that similar to what the A900 is doing below ISO 200?

When I ran tests between ISO 100 - 320 on the A900, I saw that below 200 the camera was over-exposing. If I shot at ISO 100 on the A900 and under-exposed the image, would I have more highlight recovery capability? Forgive me for bringing up this point again, but is ISO 320 simply the only way to get the greatest DR from the A900? I know I need to study your reply to one of my previous posts, but I'm really just trying to grasp the "sweet" spot for the A900's DR, as a landscape enthusiast...

[agorabasta]

One more thought on 12/14bit issue - since we continue to get more and more megapixels anyway, the need for 14bit gets smaller and smaller for the same simple reason that smaller sensels get less photons per sensel, and counting them properly takes less digital levels. E.g. an a900 could show some shadow posterisation at ISO below 320 only if those shadows were raised by over 5 stops, so moving that posterisation threshold below ISO 100 needs only about 1.7 stops less light per sensel, which means an FF sensor of about 70-80Mp would have no need for more than 12bit whatsoever.
That would mean a 40Mp APS for 12bit, 80Mp for 11bit, 160Mp for 10bit, etc. That's if the lowest is ISO100. Then if it's ISO200, a 20Mp APSC already does not need more than 12bit, and it's exactly the density of Canon 7D... So we're almost there.

[OneGuyKs]

See this thread. DR scores may not be even real
http://forums.dpreview.com/forums/readf ... d=36859818

[bakubo]

agorabasta and others might find this interesting and can comment on whether it makes any sense.

Pentax K5 High DR is just a trick:

http://forums.dpreview.com/forums/read. ... e=36859818

[agorabasta]

Henry,

If the DxO were of any credibility (or me of more credulity), there could be a matter to discuss. As it is - no way.

Still here's a more substantial comment - the divergence at lowest ISOs is real, it comes from 14bit processing. What's not real, is the nonlinearity of the K5 plot - it must stay linear while the a55 one flats out towards the lower ISO.
The K5 'trick' is no cheat, it's only natural to have higher blackpoint at higher ISO to accommodate the negative signal values from the bipolar nature of read noise. Having rectified the bipolar noise by lower blackpoint would have moved the averages, thus affecting the colour.

So there's something real about the K5, but it's not cheating on the Pentax part. And there's also some pseudoscience from DXO, as usual.

P.S. Their 'DxO Optics' ver. 6.5 blasphemy is nothing better than ver. 6/6.2. The basic demosaicing is sure much better as is everyone's else these days, yet the final result is abysmal as it's a norm with the DxO.