[lightbulb]

Hi guys,

The monkey been bugging me to upgrade to k3. Could I dial down the mp to 16 and get equally good high iso performance as my k5iis? If so I am ready to go! I need high iso to freeze action.

Thanks.

[JimmyDranox]

Will not help. But as far as I know there is no such big difference in high ISO between K-3 and K-5 IIs. But I must say that I didn't test it personally.

[philbaum]

Hi guys,

The monkey been bugging me to upgrade to k3. Could I dial down the mp to 16 and get equally good high iso performance as my k5iis? If so I am ready to go! I need high iso to freeze action.

Thanks.

I've read that the iso performance in the K3 is just as good as the k5iis. I've shot in some dark theater scenes recently at iso's above 20,000, and didn't notice any more difficulty than with my K5. Light room noise reduction controls seem to do a fairly good job. one image i shot at high iso, printed it to 20x30 :-))

But in dark scenes - the low light focus capability of the K3 is worth the price of admission. It really works, -3ev. wow!!!

[vonBaloney]

There is also really no such thing as "dialing down" the MPs -- it is a 24MP sensor, and that's that. So making that into a 16MP jpg is just throwing away data -- it doesn't get you anything other than smaller size.

[i83N]

watch this please (;

Youtu.be

Youtu.be

[carrrlangas]

There is also really no such thing as "dialing down" the MPs -- it is a 24MP sensor, and that's that. So making that into a 16MP jpg is just throwing away data -- it doesn't get you anything other than smaller size.

But it´s called downsampling, as in Sample a large ammount of data into a smaller ammount. You can shoot RAW at 24MP and then downsample to 16mp in post prossesing. You´d be sampling 1,5 pixels into 1 (and there are many technics to sample data) and noise would be square root times the sampling rate: Square root(24/16) = 1,2247..., so noise would be 22% better than the same image taken at 16Mp, (unsampled). All this should work assuming noise level of the 24Mp sensor is the same as in the 16Mp sensor at the same ISO setting.

[hcc]

In a recent thread [1], PF member bdery shared his experience [2]:
"you can shoot at 16 MP and get much improved high ISO niose control. Better than the K-5 II from my experience."
"I get better per pixel noise handling that way. [...] Print a high ISO image from the K-5 and the same from the K-3, both with the sme print size, and you'll see what I mean."

[1] https://www.pentaxforums.com/forums/172-pentax-k-3/263120-initial-review-comments-k-7-k-3-a.html
[2] https://www.pentaxforums.com/forums/172-pentax-k-3/263120-initial-review-comm...ml#post2824848)

I cannot comment myself since I do not a K-5, but I found the information an interesting technique to reduce in-camera the noise.

Personally I prefer to have more noise, more resolution and more mega-pixels (Mp), and to post-process (PP) in computer the image afterwards. More Mp some enables cropping, more resolution (ie full sensor resolution) gives me more control and there are some very good noise reduction software IMHO.

Hope that the comment may help.

[bdery]

There is also really no such thing as "dialing down" the MPs -- it is a 24MP sensor, and that's that. So making that into a 16MP jpg is just throwing away data -- it doesn't get you anything other than smaller size.

That is not true, sorry.

watch this please (;

this video has truth and falsehoods all mixed together. I wish it would stop circulating.

To answer the OP's question, yes using the K-3 at 6 MP instead of 24 (as I do when shooting JPEGS) will improve the handling of noise. The camera actually does an excellent job at downsampling, possibly because it is able to fully remap the Bayer array and take advantage of all the available information. I find that the perpixel noise is better at 16 MP than at 24 MP. However, that's misleading because each effective pixel is smaller at 24 MP, so noise per area is not really better.

[DSims]

Don't let the camera do the down-sampling if you want the highest quality. For example, on the K-5 or K-5 IIs, full resolution JPEGs (16MP) with only 1 star on the quality setting come out small in file size, but show minimal loss of IQ. Reducing the MP in-camera gives a much more noticeable loss of quality. So you can try it, but you might not like the results.

[i83N]

this video has truth and falsehoods all mixed together. I wish it would stop circulating.

this one is the new one...

[vonBaloney]

That is not true, sorry.

You'll have to elaborate on that -- the pixels sites are hardware, not software, right? (As is the bayer array -- it is built into the sensor.) So any lowering of the resolution is necessarily a software algorithm, and applied to the raw data coming from the sensor, right? (In other words, it can be done in-camera, but it can also be done later in post-processing). But the image capture is always at the full-resolution of the sensor. Unless you are arguing that when you set the camera to 16MP it is actually turning hardware pixel sites OFF (or otherwise ignoring the data from them completely), then my statement is true. (Since that was the meaning of my statement -- that the sensor always captures its full resolution and it can't magically fluidly transform itself into a sensor with bigger pixels because the hardware is the hardware.)

[maltfalc]

you'll end up with less noise, PER PIXEL, and fewer pixels, which is pointless in the real world. if the final image is going to be a web sized jpeg displayed online, it won't make any difference if you went from 24 to 16 to 1MP or straight from 24 to 1. if you're making prints you won't see any difference up to a 16MP print. with any prints larger than that the 24MP image will start to look noisier close up but will still gain additional detail over smaller prints. the 16MP image won't get any noisier, but it'll start looking soft.

[bdery]

You'll have to elaborate on that -- the pixels sites are hardware, not software, right? (As is the bayer array -- it is built into the sensor.) So any lowering of the resolution is necessarily a software algorithm, and applied to the raw data coming from the sensor, right? (In other words, it can be done in-camera, but it can also be done later in post-processing).

Yes and no. What you say is true, but since for each pixels you actually have three colour channels (sometimes it's more complicated than that, with more green photosites for instance) you can remap the sensor when lowering the resolution. With a CMOS it could potentially be different since some processing can be done on the sensor. But I believe the colour channels remain separate at that level.

but the image capture is always at the full-resolution of the sensor

Correct, at the analog level.

Unless you are arguing that when you set the camera to 16MP it is actually turning hardware pixel sites OFF

No.

Since that was the meaning of my statement -- that the sensor always captures its full resolution and it can't magically fluidly transform itself into a sensor with bigger pixels because the hardware is the hardware.

It is correct to say that the sensor operates at full resolution. But the effective pixels (remapped from the bayer array) could indeed be larger.

fewer pixels, which is pointless in the real world.

Not quite pointless. Processing is faster (case in point with lens correction on the K-3), writing to the memory card is faster per file, memory fills more slowly, transfers and post-processing are faster, etc.

[vonBaloney]

Yes and no. What you say is true, but since for each pixels you actually have three colour channels (sometimes it's more complicated than that, with more green photosites for instance) you can remap the sensor when lowering the resolution. With a CMOS it could potentially be different since some processing can be done on the sensor. But I believe the colour channels remain separate at that level.

Is there any evidence that this "remapping" is actually happening? In other words that it is doing something special that I cannot duplicate later just by downsampling the RAW data? And how exactly would this still not be "throwing away data"? (I did not expect that to be a controversial statement since 16MP is by definition less than 24MP of data.)

It is correct to say that the sensor operates at full resolution. But the effective pixels (remapped from the bayer array) could indeed be larger.

But the actual physical pixel sites are of fixed size and with a fixed color filter (R,G, or B) and that's the entire reason noise is worse with a higher resolution sensor (that is the same size as a smaller resolution one) -- the *physical* pixels are smaller and can hold less photons. Effective pixels are just a mixture of physical pixels, right? So anything that happens after capture is software trickery. So by downsampling (essentially what this remapping would be, no?) you can have less noise, yes, but you also have less detail. You have less of everything as expected (and again, by definition). That's not the same thing as "improved noise handling".

Here's the scenario that I'm skeptical of, and I'd need proof of:

-- We take a K-3 and set it to RAW+ so it gives us the RAW file and also a JPEG-- jpegs are set for 16MP
-- We take an image (high-ISO)
-- You get the 16MP jpeg and I get the 24MP RAW file as our starting points
-- We each make the best "technical" print of that image we can -- most detail / least noise. (Still subjective, I know, but what can you do?)

-- Will your print be better than I can *possibly* make mine? (i.e. you gained something magical by using the smaller resolution jpeg file that I can't do in post-processing even with the full RAW file)

I doubt it. That's a strong claim.


But more likely is that you're not claiming something magical that can only be done at capture time, but simply that downsampling as a first step from the full resolution RAW data will give you improved noise handling. But again, even if that's true (not sure that it is given same print sizes) you're throwing away detail to do it. (Which is fine because there might be plenty to spare, but still...) I mean, that's what noise reduction software does -- smooths over noise but it also smooths over detail, and you find a balance point where the noise stops distracting you without the lack of detail annoying you too much...

[maltfalc]

Not quite pointless. Processing is faster (case in point with lens correction on the K-3), writing to the memory card is faster per file, memory fills more slowly, transfers and post-processing are faster, etc.

pointless when it comes to reducing noise, which is the only point my reply was addressing, and the whole point of this thread.