[jsherman999]

....
The tradeoff in pixel noise levels on the K3 vs K5 is that we have many more pixels on the K3. So when sized and viewed similarly the apparent noise is similar.
I have the K3 and K5. In most situations its very hard to tell which photo the K3 versus the K5 took. However the resolution difference is quite noticeable.

Thanks for the info. You used Bob Newman's (bobn2, sensorsgen) methodology for the QE, correct?

The K3 looks like a very nice camera - a K5 with regard to image noise, but with more resolution, as we would expect due to the similar QE. Have you tried any comparative NR or sharpening on both raws?

.

[Stagnant]

I know these shots are just a test, but no-one would normally consider using ISO 12800 for such a night scene 'landscape' shot, especially if they were interested in detail preservation, and had access to a tripod.

IS0 200 or 400 + f5.6 + a sturdy tripod would probably be how one would normally shoot such a scene.

ISO 12800 for indoors low-light handheld, gigs in dark bars, low-light sports with telephoto, maybe OK. ISO 12800 for static night landscapes (or astrophotography), probably never OK.

These shots were meant to push the camera to the limit in very harsh conditions. As you might have noticed there are shots with more or less realistic settings (ISO 400, ISO 200 both at F9), considering the presence of a tripod.

I decided for myself that I can safely go up to 12800 in normal conditions and fix the noise in post processing. 12800 or even 6400 with K-01 was last resort, as even post processing could not compensate for the amount of noise and in-camera smoothing.

[dosdan]

I know these shots are just a test, but no-one would normally consider using ISO 12800 for such a night scene 'landscape' shot, especially if they were interested in detail preservation, and had access to a tripod. IS0 200 or 400 + f5.6 + a sturdy tripod would probably be how one would normally shoot such a scene.

If you look at the other shots in his folder, you'll see he does use a low ISO. I thought IMGP7188 | Flickr - Photo Sharing! taken at ISO100 with a K-01 turned out well.

Regarding the optimum settings for shooting the scene in question: the settings used, 0.3s, f/9, ISO12800 indicates a Light Value of 1.1 LV. While Stagnant mentioned that a gust of wind stuffed up his ISO100 shoot, if the wind was still enough you'd want to use base ISO. (I've had great difficulty myself shooting long exposure shots on a windy night.) Suitable settings for this LV, without use Bulb mode, would be 30s, f/9, ISO125 or 30s, f/8, ISO100.

This was shot with the kit lens set to 50mm FL, so stopping down to f/9 was a good idea. If shooting with a 50/F1.4 lens, and DOF was not an issue, (and it wasn't so foggy), I've have used f/4-f/5.6 here i.e. 8s, f/4, ISO100 or 15s, f/5.6, ISO100.

But let's say, that due to the wind, you didn't want to go slower than 0.3s shutter speed. I think the shot, as rendered, is blown too much. (As mentioned he did not make any adjustments.) But the in-camera highlight blowing is ADC-clipping rather than FWC saturation, since 128x of digital gain is being applied. So instead of shooting at ISO12800, for the same exposure settings (the same Scene Luminance, Shutter Speed, f-stop), I'd have shoot at ISO1600 (the last stop to use analogue gain), and boosted 3 stops (8x) in PP raw development to get to an equiv. ISO12800 rendering brightness level, while playing with the top of the Tone Response Curve to prevent the Digital Number maxing out. (Although, I'd probably end up boosting it less than 3 stops anyway.) This would have given 3 stops more headroom during the capturing phase. If you were to just apply 3 stops of boost in PP then you'd still end up with the same blown highlights. But since you can still alter the shape of the TRC at this time, you can still make adjustments so the highlight is rendered below the max. DN limit.

Dan.

---------- Post added 08-03-14 at 08:51 ----------

Stagnant, I've downloaded your PEF. Thanks for making it available. I've got an athletics meet to shoot today so I'm a bit busy, but when I get time, I'll make up a SP 5, RT, DCU 5 composite comparison image and present it here. (I don't have LR.) That will confirm your WB & noise performance results, and add RawTherapee into the mix.

Dan.

[Rondec]

Rondec, would you mind, uploading the original image size, or at least a 100% crop ?

This is a roughly 100 percent crop of the above photo, no noise reduction.

[Quicksand]

This is a roughly 100 percent crop of the above photo, no noise reduction.

Sorry Vincent, I hate to be "that guy," but first of all . . . that's ISO 4000 per flickr, not 6400.

But even so, I have a hard time believing that there is NO noise reduction. The total lack of chroma noise just simply isn't consistent with that.

It's possible that you're not intentionally applying any NR, but I think there has to be some being applied somewhere in the process, either in-camera, or via some default baseline settings in development software.

Frankly, that looks about like ISO 800 on my K-5.

[Rondec]

Sorry Vincent, I hate to be "that guy," but first of all . . . that's ISO 4000 per flickr, not 6400.

But even so, I have a hard time believing that there is NO noise reduction. The total lack of chroma noise just simply isn't consistent with that.

It's possible that you're not intentionally applying any NR, but I think there has to be some being applied somewhere in the process, either in-camera, or via some default baseline settings in development software.

Frankly, that looks about like ISO 800 on my K-5.

Well, thanks for that.

You are right that it is iso 4000. As to whether it looks like iso 800 on your K5, I have no idea, but it doesn't look like iso 800 on my K5 II. This is why I hate 100 percent crops. They have nothing to do with real life printing, viewing or shooting. A 100 percent crop on a K3 is a lot tighter than what you could on a K5 and so the comparison is not valid. Upsample your iso 4000 K5 image to 24 megapixel equivalent and then 100 percent crop it and post it and then I'll believe it. As to noise reduction, I think Lightroom must do some noise reduction that I didn't uncheck, although I don't usually mess around with it either way. So sorry if that spoiled my image.

Edit: Sorry. I was a little irritated. I am not someone who posts comparisons between cameras, because I know I can't do them "right" and control all parameters, focus points, etc in such a way as to make these comparisons valid. That said, for what it is worth, with regard to real life shooting, I find my K3 to be equivalent to my K5 II with regard to high iso shooting and to have more resolution at lower isos, but slightly less dynamic range at base iso.

[Fauxton]

Thanks for the info. You used Bob Newman's (bobn2, sensorsgen) methodology for the QE, correct?

The K3 looks like a very nice camera - a K5 with regard to image noise, but with more resolution, as we would expect due to the similar QE. Have you tried any comparative NR or sharpening on both raws?

.

es that is Bobn2 and DSPographers method.

Yes I've look at the sharpening quite closely. I've tested a couple of methods on the raw files of both cameras. I find not all sharpening is the same. Here is what I've found:
DCU 5 sharpening is quite good with either camera however with the K3 it can sharpen quite strongly without halo's. I don't like the interface particularly and I rarely use DCU. But it can provide good results. It is particularly good for dialing in the camera JPEG settings as you can test what settings work and them set them in the camera.
DXO 9 uses its lens software algorithms very well on the K5 and K3. I've used them against a variety of lens and the effects are very good. No halos on either the K3 or K5 but on the K3 I can use stronger sharpening without fear of making it overly crunchy. The extra resolution of the K3 is very apparent when using DXO9 relative to what can be done with the K5.
Once I've used DXO I typically work the keepers further in CS6. Some things I may do are use the unsharp mask (high radius, low amount) for microcontrast and sometimes I may use hi pass blending or smart sharpen depending on what effect I'm after. The K3 can take this sort of thing better than the K5. (I don't use all three together)
I've had very good results with Nik plugins and the K3 responds very nicely to both the sharpener and the noise reduction software. Good stuff there.
Perfectly Clear does a very good job on both the K5 and K3 files. The results are very good on the K3 at its default settings and I find I don't need to tune that filter much. It sharpens the photos nicely without halos. This plugin is simply great.
I've also tried Raw Therapee with similar results. So the bottom line I find is this:

I can sharpen K3 files further than I can K5 and maintain the integrity of the image better. For my workflow and what I'm looking for noise is not a concern up to 1600 on either. Past that and I find the noise similar with the K3 degrading a bit quicker (on the K3) if the pixels are peeped at 100% for both. I put that down to the inherent NR being applied on the K5 - there is default NR on the K3 but it is clearly less. Note that if the images are sized so that the image appears magnified the same - apparent magnification then the results for noise are similar but the detail on the K3 is much better. Once NR software is applied at 3200 and above the result is similar however I find that the K3 can still retain more detail despite the increased application of NR over the K5 at 3200 and above.
One thing that I find I need to watch with the K3 is the suppression of the reds in the photograph when addressing color noise. More NR is required on the K3 than the K5 and this can cause some color suppression if one is not careful.

As to read noise on the K5 and K3 similar read noise. Sensorgen calculated the read noise of the K5 at 1600 is 1.9 and I calculate the K3's read noise at 2.3 at 1600. The delta likely would be inside the manufacturing, measurement and calculation error of the method. I've calculated the QE for the K5 at 46% and the K3 at 48% which again is pretty much a wash. What is different is the Full Well Saturation. The K5 is higher than the K3 hence its better DR. This higher saturation provides the K5 benefits that the K3 offsets at higher ISO with trading the SNR for resolution (it has a fair amount it can trade for example Imaging Resource estimated the K5 resolution on a test at 2100 lines and the k3 on a similar test with similar method of interpretation as 2700 lines - so it has about 28% more resolution on THAT test - YMMV) - that works to a point and for me that cross over appears to be about 6400 depending on the content of the picture. But even then we are only talking about a half stop of DR at the same ISO between two class leading cameras. For me the gains of the the K3 resolution more than addresses the better Full Well Saturation point of the K5 particularly at the light levels and lenses I intend to use it with. I'm of the opinion that the K3 sensor is pretty much using the same design and manufacturing process as the 16 MP sensor. The difference is in the size of the pixels and their inherent FWC. We did not get anything new per se we just got more of it.

But I do like the results and all the calculations provide is comfort that the K3 is (as per my observation and again YMMV) indeed providing better IQ than the K5, and up to 3200 clearly better.

[Parry]

es that is Bobn2 and DSPographers method.

Yes I've look at the sharpening quite closely. I've tested a couple of methods on the raw files of both cameras. I find not all sharpening is the same. Here is what I've found:
DCU 5 sharpening is quite good with either camera however with the K3 it can sharpen quite strongly without halo's. I don't like the interface particularly and I rarely use DCU. But it can provide good results. It is particularly good for dialing in the camera JPEG settings as you can test what settings work and them set them in the camera.
DXO 9 uses its lens software algorithms very well on the K5 and K3. I've used them against a variety of lens and the effects are very good. No halos on either the K3 or K5 but on the K3 I can use stronger sharpening without fear of making it overly crunchy. The extra resolution of the K3 is very apparent when using DXO9 relative to what can be done with the K5.
Once I've used DXO I typically work the keepers further in CS6. Some things I may do are use the unsharp mask (high radius, low amount) for microcontrast and sometimes I may use hi pass blending or smart sharpen depending on what effect I'm after. The K3 can take this sort of thing better than the K5. (I don't use all three together)
I've had very good results with Nik plugins and the K3 responds very nicely to both the sharpener and the noise reduction software. Good stuff there.
Perfectly Clear does a very good job on both the K5 and K3 files. The results are very good on the K3 at its default settings and I find I don't need to tune that filter much. It sharpens the photos nicely without halos. This plugin is simply great.
I've also tried Raw Therapee with similar results. So the bottom line I find is this:

I can sharpen K3 files further than I can K5 and maintain the integrity of the image better. For my workflow and what I'm looking for noise is not a concern up to 1600 on either. Past that and I find the noise similar with the K3 degrading a bit quicker (on the K3) if the pixels are peeped at 100% for both. I put that down to the inherent NR being applied on the K5 - there is default NR on the K3 but it is clearly less. Note that if the images are sized so that the image appears magnified the same - apparent magnification then the results for noise are similar but the detail on the K3 is much better. Once NR software is applied at 3200 and above the result is similar however I find that the K3 can still retain more detail despite the increased application of NR over the K5 at 3200 and above.
One thing that I find I need to watch with the K3 is the suppression of the reds in the photograph when addressing color noise. More NR is required on the K3 than the K5 and this can cause some color suppression if one is not careful.

As to read noise on the K5 and K3 similar read noise. Sensorgen calculated the read noise of the K5 at 1600 is 1.9 and I calculate the K3's read noise at 2.3 at 1600. The delta likely would be inside the manufacturing, measurement and calculation error of the method. I've calculated the QE for the K5 at 46% and the K3 at 48% which again is pretty much a wash. What is different is the Full Well Saturation. The K5 is higher than the K3 hence its better DR. This higher saturation provides the K5 benefits that the K3 offsets at higher ISO with trading the SNR for resolution (it has a fair amount it can trade for example Imaging Resource estimated the K5 resolution on a test at 2100 lines and the k3 on a similar test with similar method of interpretation as 2700 lines - so it has about 28% more resolution on THAT test - YMMV) - that works to a point and for me that cross over appears to be about 6400 depending on the content of the picture. But even then we are only talking about a half stop of DR at the same ISO between two class leading cameras. For me the gains of the the K3 resolution more than addresses the better Full Well Saturation point of the K5 particularly at the light levels and lenses I intend to use it with. I'm of the opinion that the K3 sensor is pretty much using the same design and manufacturing process as the 16 MP sensor. The difference is in the size of the pixels and their inherent FWC. We did not get anything new per se we just got more of it.

But I do like the results and all the calculations provide is comfort that the K3 is (as per my observation and again YMMV) indeed providing better IQ than the K5, and up to 3200 clearly better.

Top banana!!!

[jsherman999]

es that is Bobn2 and DSPographers method.

Yes I've look at the sharpening quite closely. I've tested a couple of methods on the raw files of both cameras. I find not all sharpening is the same. Here is what I've found:
DCU 5 sharpening is quite good with either camera however with the K3 it can sharpen quite strongly without halo's. I don't like the interface particularly and I rarely use DCU. But it can provide good results. It is particularly good for dialing in the camera JPEG settings as you can test what settings work and them set them in the camera.
DXO 9 uses its lens software algorithms very well on the K5 and K3. I've used them against a variety of lens and the effects are very good. No halos on either the K3 or K5 but on the K3 I can use stronger sharpening without fear of making it overly crunchy. The extra resolution of the K3 is very apparent when using DXO9 relative to what can be done with the K5.
Once I've used DXO I typically work the keepers further in CS6. Some things I may do are use the unsharp mask (high radius, low amount) for microcontrast and sometimes I may use hi pass blending or smart sharpen depending on what effect I'm after. The K3 can take this sort of thing better than the K5. (I don't use all three together)
I've had very good results with Nik plugins and the K3 responds very nicely to both the sharpener and the noise reduction software. Good stuff there.
Perfectly Clear does a very good job on both the K5 and K3 files. The results are very good on the K3 at its default settings and I find I don't need to tune that filter much. It sharpens the photos nicely without halos. This plugin is simply great.
I've also tried Raw Therapee with similar results. So the bottom line I find is this:

I can sharpen K3 files further than I can K5 and maintain the integrity of the image better. For my workflow and what I'm looking for noise is not a concern up to 1600 on either. Past that and I find the noise similar with the K3 degrading a bit quicker (on the K3) if the pixels are peeped at 100% for both. I put that down to the inherent NR being applied on the K5 - there is default NR on the K3 but it is clearly less. Note that if the images are sized so that the image appears magnified the same - apparent magnification then the results for noise are similar but the detail on the K3 is much better. Once NR software is applied at 3200 and above the result is similar however I find that the K3 can still retain more detail despite the increased application of NR over the K5 at 3200 and above.
One thing that I find I need to watch with the K3 is the suppression of the reds in the photograph when addressing color noise. More NR is required on the K3 than the K5 and this can cause some color suppression if one is not careful.

As to read noise on the K5 and K3 similar read noise. Sensorgen calculated the read noise of the K5 at 1600 is 1.9 and I calculate the K3's read noise at 2.3 at 1600. The delta likely would be inside the manufacturing, measurement and calculation error of the method. I've calculated the QE for the K5 at 46% and the K3 at 48% which again is pretty much a wash. What is different is the Full Well Saturation. The K5 is higher than the K3 hence its better DR. This higher saturation provides the K5 benefits that the K3 offsets at higher ISO with trading the SNR for resolution (it has a fair amount it can trade for example Imaging Resource estimated the K5 resolution on a test at 2100 lines and the k3 on a similar test with similar method of interpretation as 2700 lines - so it has about 28% more resolution on THAT test - YMMV) - that works to a point and for me that cross over appears to be about 6400 depending on the content of the picture. But even then we are only talking about a half stop of DR at the same ISO between two class leading cameras. For me the gains of the the K3 resolution more than addresses the better Full Well Saturation point of the K5 particularly at the light levels and lenses I intend to use it with. I'm of the opinion that the K3 sensor is pretty much using the same design and manufacturing process as the 16 MP sensor. The difference is in the size of the pixels and their inherent FWC. We did not get anything new per se we just got more of it.

But I do like the results and all the calculations provide is comfort that the K3 is (as per my observation and again YMMV) indeed providing better IQ than the K5, and up to 3200 clearly better.

Excellent, thanks very much for sharing your efforts.

[dosdan]

Here's my processing of Stagnant's PEF with Silkypix Developer Studio 5, RawTherapee 4.0.12.60, Pentax Digital Camera Utility 5.0.

First off, SP is difficult for use this comparison as it. doesn't really have a neutral setup. The global "Default" taste makes use of the sensor type (CCS/CMOS), pixel pitch, resolution and ISO to set a slew of parameters.

Also SP 5 defaults to a different colour taste, Natural, whereas SP 4 used a set of tastes, some of which are now included in SP 5 as Faithful versions. As you can see below, the default is now a bit more red-yellowish:




Note: all 2-panel comparisons, were made from screen-grabs with a SP zoom setting of 87%. Then they were resized from 1,400px wide down to 1,024px wide. This means that the version shown here is 63% of the original width & height, and 40% of the original area. This downsizing should give a 4dB improvement to the SNR.

Also the fault contrast taste is: Average Contrast, whereas Standard Constrast is more typical of other programs:



The "No Sharpness" taste in SP 5 affects both the Sharpening settings & the NR settings. (Because these are so closely linked, tastes for this part of the program record the settings used in both sections.) However, while sharpening is fully zeroed, NR with this taste still has the False Colour Controller set to 30. This reduces the noise somewhat:




With these initial differences mentioned, here is a 100% crop which I've combined from the 3 programs without resizing:



I've used the SP settings which seemed to better match RT's colour. As you can see there is high-freq noise in RT, but not in SP which mainly has low-freq noise. DCU 5 obviously has major chroma NR, but is using less luma NR, and this is with all NR options unticked.

Finally, I let SP 5 use make use of the processing defaults it chose for this image. I think it does a decent job for an ISO12800 image. The only part I thought that might be a bit over-smoothed is the vertical girder assembly on the left-hand side, but this is not that visible either on the version without NR.




Dan.

[Stagnant]

At this point it is seems that the performance of sensor gives you only a half of the result, the other half depends on the post processing. It also seems, that each converter even at default or zeroed settings gives different results, so it is hard to decide what the contents of the raw file actually is like.
dosdan , thanks for the comparison. If you shoot K-3 in DNG, in SilkyPix you have an option to choose "DNG" instead of "Natural" or "Faithful". In case with DCU vs RT vs SP, I would say that RT produced a more detailed image.

[Mr Spocko]

I suspect some under the hood RAW NR is going on too, at least it has on every recent Pentax I've used.

[Stagnant]

According to some website (I can't remember the name), measurements showed that there is no RAW NR in K-3. It certainly is present in K-01.

[Parry]

ISO800 with the K-3 looking through the Meyer Optik Gorlitz Orestegor 300/4 . . . at f/11.



Oh and this is cropped.

[jsherman999]

ISO800 with the K-3 looking through the Meyer Optik Gorlitz Orestegor 300/4 . . . at f/11.



Oh and this is cropped.

Gotta say, K3 looks like a killer choice.