[Laurentiu Cristofor]

I was checking the dxomark test results when I noticed that the K-x performance is getting very close and in some cases even gets better than that of the K-5 at high ISO.

Basically, starting with ISO 3200, the K-x graph starts getting closer to the K-5 one and in the case of dynamic range even surpasses the performance of the K-5 at ISO 12800 (despite a larger gap at lower ISOs).

Pretty interesting behavior.

[rawr]

recent & related:

https://www.pentaxforums.com/forums/pentax-dslr-discussion/168366-k-x-vs-k-5-...so-rating.html

But yes, the K-x does still punch above it's weight. However I am finding detail retention in the K-5 exceeds that of K-x high-ISO's.

[dosdan]

The reason why the K-x DR curve intersets the K-5 at high ISO is the more aggressive level of raw NR applied. Notice how much the DR curve slope changes. This means, MP differences aside, that it is likely that the K-x will have less detail at high ISO than the K-5.

Dan

[Laurentiu Cristofor]

recent & related:

https://www.pentaxforums.com/forums/pentax-dslr-discussion/168366-k-x-vs-k-5-...so-rating.html

Yes, I remember that thread. However, that one started with dpreview's comparison and while dxomark was mentioned briefly, no one pointed out that their results actually support those of dpreview.

The reason why the K-x DR curve intersets the K-5 at high ISO is the more aggressive level of raw NR applied. Notice how much the DR curve slope changes. This means, MP differences aside, that it is likely that the K-x will have less detail at high ISO than the K-5.

The curves always change when the smoothing occurs, but it's interesting that the K-x one manages to overcome the K-5 one. And since this is the DR curve, I'm not sure we can say that the K-5 can retain more detail if it actually has less DR - it may retain some fine detail, but it may lose detail in shadows and highlights.

[dosdan]

The curves always change when the smoothing occurs, but it's interesting that the K-x one manages to overcome the K-5 one.

And since this is the DR curve, I'm not sure we can say that the K-5 can retain more detail if it actually has less DR - it may retain some fine detail, but it may lose detail in shadows and highlights.

Laurentiu, why do you think the DR curve changes when smoothing is applied?

DR is FWC-to-noise-floor ratio with the noise floor made up of ADC noise, noise from the ISO sensitivity programmable gain amplifier (PGA) stage, sensor read output noise & photon/shot noise. (The reason for the change in DR vs ISO is covered in this post: https://www.pentaxforums.com/forums/pentax-k-5-forum/162940-k5-vs-k7-dynamic-...ml#post1689551)

Once "smoothing" (raw NR) kicks in, the image DR decreases at a much low rate than expected. This is because the total read noise, after NR, is less abundant. But NR and sharpness are two sides of the same coin, as one goes up, the other comes down. So how are you going to get both smoothing through raw NR and yet maintain fine detail, in the "smoothed" shadows? The answer is, "You won't".

Smoothing/NR (less noise) = less high frequency component in the image (HF in an image is biggest at abrupt transitions, like sharp edges) = less detail.

Dan.

[Laurentiu Cristofor]

(The reason for the change in DR vs ISO is covered in this post: K5 vs k7 dynamic range)

That looks like the kind of information I always wanted to know, but I'll need some time to digest it.

So how are you going to get both smoothing through raw NR and yet maintain fine detail, in the "smoothed" shadows? The answer is, "You won't".

I guess what I am missing is this: ok, there is less fine detail in the smoothed shadows/highlights, but at least there is some, vs. the alternative of clipping them. I guess you might be saying that the loss of detail to clipping is more negligible compared to the loss of detail elsewhere due to NR? Is that it?

[dosdan]

I guess you might be saying that the loss of detail to clipping is more negligible compared to the loss of detail elsewhere due to NR? Is that it?

I wasn't talking about clipping at all. In the studio where everything is controlled, you can exposure right up to just before the clipping point. In the field, it's better to keep away from it, as the output becomes unpredictable and blown highlights are a good way to ruin a shot.

This camera does not respond well to strong ETTR or going past it. (There is less need anymore for this behaviour in an attempt to shoe-horn an image's DR into a camera's limited DR.) It has been calibrated so you really want to stay away from bouncing against the right of the luminance histogram. (Some cameras are calibrated so there is still a bit left in the tank so they're more forgiving of reaching or slightly exceeding the right of the histogram.) With this camera, the noise floor is so far down, it's better to underexpose up to a stop, and maintain some clipping headroom without having to worry too much about noisy shadows.

With the K-5, the best strategy is only ever use up to ISO1600, shoot raw, underexpose instead of raising ISO above that (all higher ISO is straight digital manipulation - like in most DSLRs) so you're losing clipping headroom unnecessarily if you're using higher ISO (unless you want to shoot JPEG), and boost the exposure in PP in your raw converter afterwards. This way you're preserving more clipping headroom, and you also avoid raw NR completely and thus have more flexibility to apply what type and amount of NR you want in PP.

Dan.

---

Last edited by dosdan; 12-31-2011 at 12:37 AM.

[Rondec]

Clearly something changes at iso 3200 for the kx, because its curve is following a smooth path down and then, it jumps and has almost the same dynamic range iso 3200 as at iso 1600. This is quite unlike the K5 curve where the path is smoothly downward throughout the various isos. To me, this says that the RAW in the kx is being cooked considerably more than the K5 starting between iso 1600 and iso 3200. How much that difference that translates into real life shooting I don't know, but I can imagine that there would be more detail with the K5 at high isos (over 3200) than with the kx.

[Laurentiu Cristofor]

I wasn't talking about clipping at all. In the studio where everything is controlled, you can exposure right up to just before the clipping point.

I think we're talking about different things then. In practice, I rarely go over ISO 800 and I only once used ISO 3200, more out of curiosity than anything else - ISO 1600 is my hard limit.

But what I was asking was this: if the K-x shows better DR at ISO 12800, isn't that a positive thing vs. the K-5 at 12800? Everything else comes about the same in the other graphs.

[dosdan]

if the K-x shows better DR at ISO 12800, isn't that a positive thing vs. the K-5 at 12800? Everything else comes about the same in the other graphs.

The DR chart indicates the noise is less than the K-5 at very high ISOs. But DR, is not the only important image metric. What the graph doesn't show is the resolution at a DR value, particularly the high ones. If you've applied heavier NR above ISO1600, yes the noise will be lower at very high ISO, but you may not like the quality of that's shown.

Dan

[UnknownVT]

Yes, I remember that thread. However, that one started with dpreview's comparison and while dxomark was mentioned briefly, no one pointed out that their results actually support those of dpreview.

Photography is a visual media -
and more often than not our eyes are the final judge -
but I will take the point that one's own judgement may not necessarily apply to anyone else.

I like displaying these from dpReview.com on the Pentax K-5 -
ISO6400:

I find ISO6400 is about the highest practical limit for noise.

RAW:


ISO12800:


RAW:

[Laurentiu Cristofor]

I like displaying these from dpReview.com on the Pentax K-5 -

A problem with such shots is that they are taken at different times and a small difference in focusing can mean a lot when comparing 100% crops.

[UnknownVT]

A problem with such shots is that they are taken at different times and a small difference in focusing can mean a lot when comparing 100% crops.

If we follow this logic to its ultimate conclusion -
then it doesn't matter even if we took the test photos ourselves with whatever utmost care -
because they are taken with two different camera bodies and so "separately" -
so there will be the possibility (no matter how small) of focus differences -
so absolutely NO comparison test photos can possibly be valid.....
huh?

Happy new year everyone
(now that we've cast doubts on everything we believed in
next: white is black
and the non-existence of b&w photography... )

[Laurentiu Cristofor]

then it doesn't matter even if we took the test photos ourselves with whatever utmost care -

No. The problem is that the dpreview test shots are taken with *NO* care for maintaining the same focusing across tests. I don't know if they even use manual focus or they let the AF system decide where to focus. I don't think they maintain the same focusing even within a test for different ISO shots - much less for different tests done weeks or months apart.

If you cared, you could set down the tripod at a fixed distance from the scene, place the two cameras on it, and take two shots while focusing in the same place - using live view for example to ensure focusing precision. Then you could compare shots focused the same way and you could pick your crops from the in-focus areas.

[UnknownVT]

No. The problem is that the dpreview test shots are taken with *NO* care for maintaining the same focusing across tests. I don't know if they even use manual focus or they let the AF system decide where to focus. I don't think they maintain the same focusing even within a test for different ISO shots - much less for different tests done weeks or months apart.

I'm sorry, I did not know that -

Would be very grateful for references on this