Shot around at 1600-3200 today to see what I could get. When properly exposed this things totally knocks my socks off!

Enjoy!

ISO 3200


ISO 3200


ISO 3200


ISO 3200


ISO 1600


ISO 1600

[codiac2600]

you're making it hard not to order one. sigh.

Sorry

Boy Chris this is some nice work. I am wondering any PP work done here? That would be the tell tale sign for me. Meaning 3200 could look good with some great PP skills. You have those skills we have seen. Clarence, I am with you about giving up lunc for awhile. Even thought about challenging my self to if I lose 50 lbs I get the K20 or DA300. Figured this would be a win, win. Plus I may even be able to convince my wife it is a good idea

The only PP is converting from RAW. There shouldn't be any noise reduction unless I left it on in ACR somehow (we all know how effective that noise reduction is too), but for the most part this is 100% no PP.

I actually prefer this grain and love the look. I'll probably use it more often than not when shooting street photography around sundown time.

[Jonson PL]

I didn't think the K20 high Iso shots were gonna look this good. Very useable, good work on Pentax/Samsung with the sensor

[Marc Langille]

Very, very nice.... hmmm... low light wildlife photos would be awesome with this level of noise.... what to do??

[falconeye]

The only PP is converting from RAW. There shouldn't be any noise reduction unless I left it on in ACR somehow (we all know how effective that noise reduction is too), but for the most part this is 100% no PP.

Nice pictures. Thanks.

However, I see it over and over again and here as well: People compare the ISO quality on images which have been resampled, either intentionally or by the upload process.

There is a simple statistical law which teaches us that a resampling to 50% size is equivalent to a reduction in ISO of 2EV, like from ISO1600 to ISO400. This is because 4 pixels are resampled to a single one which got 4x as much light, corresponding to 2EV.

BTW. This is why the K20D already has better ISO performance by definition: In order to compare to K10D, images must be resampled to ~82% size which reduces noise in the process.

The first image in this thread is 1024 x 680 at ISO3200. It could be a crop but let's assume that it isn't. It is then resampled to 22%. ISO effective = ISO3200 * 0.22 * 0.22 = ISO150.

It is therefore a trivial statement that ISO3200 images, resampled to ~25% size, show no visible noise anymore.

So this entire thread has great shots but cannot make a statement whatsoever about the K20D's high ISO performance.

BTW, I hope the flower picture is a crop because it has slight visible noise.


@Codiac: could you please confirm/contradict those are resampled images and do you have original size crops? Thank You. And no offense intended. I really like your work.

---

Last edited by falconeye; 03-03-2008 at 07:26 AM.

[codiac2600]

Nice pictures. Thanks.

However, I see it over and over again and here as well: People compare the ISO quality on images which have been resampled, either intentionally or by the upload process.

There is a simple statistical law which teaches us that a resampling to 50% size is equivalent to a reduction in ISO of 2EV, like from ISO1600 to ISO400. This is because 4 pixels are resampled to a single one which got 4x as much light, corresponding to 2EV.

BTW. This is why the K20D already has better ISO performance by definition: In order to compare to K10D, images must be resampled to ~82% size which reduces noise in the process.

The first image in this thread is 1024 x 680 at ISO3200. It could be a crop but let's assume that it isn't. It is then resampled to 22%. ISO effective = ISO3200 * 0.22 * 0.22 = ISO150.

It is therefore a trivial statement that ISO3200 images, resampled to ~25% size, show no visible noise anymore.

So this entire thread has great shots but cannot make a statement whatsoever about the K20D's high ISO performance.

BTW, I hope the flower picture is a crop because it has slight visible noise.


@Codiac: could you please confirm/contradict those are resampled images and do you have original size crops? Thank You. And no offense intended. I really like your work.

You can actually draw conclusions from these images and the jargon of shrinking the image size will not make this look like ISO 150. if I shot ISO 150 and showed the same image they wont look alike at all, even at 100% crops.

Again, how many people have shown off ISO 1600 images from the K10D this large? Not very many cause they looked like crap most of the time or were converted to black and white.

The flower image was taken in the shade, complete shade.

[Heinrich Lohmann]

Nice pictures. Thanks.

However, I see it over and over again and here as well: People compare the ISO quality on images which have been re-sampled, either intentionally or by the upload process.

There is a simple statistical law which teaches us that a resampling to 50% size is equivalent to a reduction in ISO of 2EV, like from ISO1600 to ISO400. This is because 4 pixels are resampled to a single one which got 4x as much light, corresponding to 2EV.

BTW. This is why the K20D already has better ISO performance by definition: In order to compare to K10D, images must be resampled to ~82% size which reduces noise in the process.

The first image in this thread is 1024 x 680 at ISO3200. It could be a crop but let's assume that it isn't. It is then resampled to 22%. ISO effective = ISO3200 * 0.22 * 0.22 = ISO150.

It is therefore a trivial statement that ISO3200 images, resampled to ~25% size, show no visible noise anymore.

So this entire thread has great shots but cannot make a statement whatsoever about the K20D's high ISO performance.

BTW, I hope the flower picture is a crop because it has slight visible noise.


@Codiac: could you please confirm/contradict those are resampled images and do you have original size crops? Thank You. And no offense intended. I really like your work.

falconeye, if what you say is a fact I should be able to take a noisy iso 1600 image (3892x 2612) downsample it to lets say 5x7" @ 240dpi and print it clean as a whistle. In my experience that is not going to happen without noise reduction of some sort.

Heinrich

[davemdsn]

Dear falconeye, can you explain to me then how it is that my 11x14 prints are so clean from the K20? What mathematical formula do you offer to explain a print that actually represents the original subject at larger than life size and does it with no grain? In fact, there is grain in the image when viewed at 100% on screen, but no noise when printed to wall hanging size.

I'm waiting.....

[falconeye]

if I shot ISO 150 and showed the same image they wont look alike at all, even at 100% crops.

Do it, you will be suprised. Do ISO150 at 100% crop and ISO3200 at 22% and you will be surprised. If the camera's noise handling isn't messed up and the scaling is done correctly (bicubic at least).

Again, how many people have shown off ISO 1600 images from the K10D this large?

I don't own a K10D. If you post a K10D image scaled to 26% or so (to compensate for #pixels) which shows noise then you have a point. And if it as simple as to say that more pixels mean larger print size.

I should be able to take a noisy iso 1600 image (3892x 2612) downsample it to lets say 5x7" @ 240dpi and print it clean as a whistle. In my experience that is not going to happen without noise reduction of some sort.

If you print at 240dpi resolution (with a K10D), you would downsample ISO1600 to ISO350 indeed. But this isn't enough to not keep some visible noise. Esp. if the original was bad on noise.

how it is that my 11x14 prints are so clean from the K20? What mathematical formula do you offer to explain a print that actually represents the original subject at larger than life size and does it with no grain? In fact, there is grain in the image when viewed at 100% on screen, but no noise when printed to wall hanging size.

I'm waiting.....

Oh, that is an easy one

I never said the K20D wouldn't have excellent noise performance. Your wall print example makes a much better point about it than codiac's web images, though.

As for the grain on screen, not on wall... It all depends on printing resolution and dithering algorithms. Printing @ 280 dpi would print the pixels 1:1. Now, maybe your wall print only used 240dpi (reducing ISO400 to ISO300) or it renders less contrast for small color variations than your screen (gamma) or the printer's dithering engine includes a noise filter. Nothing to be too surprised about. Be happy it isn't the other way round BTW, what printing service did you use?

[davemdsn]

falconeye, you seem to be missing the point. Or maybe dodging it. You have complained in several threads that the K20 noise handling can't be determined by web size images. I agree that small screen images don't show much of anything, however, we have the testimony of a few people that the images are printing with the same or even better results as these small samples. I am not talking about a 400 ISO print at 11x14, I am talking about a 1600 ISO print at 11x14. Even the out of focus shadows show no grain. The image quality isn't passible, it's phenomenal.

You have done nothing but whine about something you haven't even seen.

[codiac2600]

Do it, you will be suprised. Do ISO150 at 100% crop and ISO3200 at 22% and you will be surprised. If the camera's noise handling isn't messed up and the scaling is done correctly (bicubic at least).




I don't own a K10D. If you post a K10D image scaled to 26% or so (to compensate for #pixels) which shows noise then you have a point. And if it as simple as to say that more pixels mean larger print size.




If you print at 240dpi resolution (with a K10D), you would downsample ISO1600 to ISO350 indeed. But this isn't enough to not keep some visible noise. Esp. if the original was bad on noise.




Oh, that is an easy one

I never said the K20D wouldn't have excellent noise performance. Your wall print example makes a much better point about it than codiac's web images, though.

As for the grain on screen, not on wall... It all depends on printing resolution and dithering algorithms. Printing @ 280 dpi would print the pixels 1:1. Now, maybe your wall print only used 240dpi (reducing ISO400 to ISO300) or it renders less contrast for small color variations than your screen (gamma) or the printer's dithering engine includes a noise filter. Nothing to be too surprised about. Be happy it isn't the other way round BTW, what printing service did you use?

Your math holds absolutely no water IMO. I don't resize any of my images anyway, just copy the images off the website that hosts my images and one size is X3 which is what I've shown here. You also forget another piece of the puzzle and that is every monitor handles noise differently. Every browser interprets images differently. The only scientific way to look at an image is on print. BTW, lets see your math take a picture?

[falconeye]

Your math holds absolutely no water IMO. I don't resize any of my images anyway, just copy the images off the website that hosts my images and one size is X3 which is what I've shown here. You also forget another piece of the puzzle and that is every monitor handles noise differently. Every browser interprets images differently. The only scientific way to look at an image is on print. BTW, lets see your math take a picture?

Codiac, this reply is a little bit more friendly. Sorry about my harsh tone in the other thread, then.

All you say is correct but has nothing to do with my argument.

It doesn't matter how you got your images down-sampled. If it was a webserver, so what?

Because everybody here seem to really have a problem to understand this simple fact -- and I am not doing math here (just take it as a fact if you don't like math) -- is this:


14.6 MPixel images, as soon as downsampled to fit a web page (1024 wide at most), do not have visible noise by definition (if not pushed for exposure, contrast, cropped etc.).

And yes, my math does take pictures! It says: If you are only after images for the web, don't care about the ISO setting (say, up to 1600).

[m8o]

falconeye... ironc handle. I have Wildabeast eyes (poor eyesight is one reason they travel with Zebras) and I see the averaged noise artifacts of a resampled image. If it's a noisy pic to begin with, I say the noise is reduced as it is averaged but evident still to any discerning viewer. Yes, 'pushing' it if it's a noisy picture to begin with amplifies it.

[falconeye]

I say the noise is reduced as it is averaged but evident still to any discerning viewer.

Ok. I accept worst preconditions.

I took out my DS (what a wonderful compact DSLR after having played with my new K20D/grip "bomb" . And I took two images in low light, one at ISO200 and one at ISO1600. Well, the DS is 3 years behind in technology and isn't exactly known for splendid ISO performance, but ok...

And I underexposed on purpose so that one has a chance to see a difference at all (in the underexposed regions). Remember my disclaimer about images not being pushed for underexposure? I even provide for a dark area bokeh so noise is easy to spot.

I'll post the images at the 1024 resolution discussed here, even if this represents a much smaller shift in ISO because of the smaller resolution of the DS.

And this is about as bad as it can be for images displayed on the web, for any current DSLR.


So, which image is which?

...and...

your discerning viewer votes, please

[janver]

Including full Exif would have been nice...

Jan V.

[falconeye]

Including full Exif would have been nice...

Jan V.

I knew somebody would try to peek the EXIF look at the images instead

One is at ISO200, the other at ISO1600. With a DS. Should be easy to spot, not?

(Note: I'll replace the images in a couple of days for versions with full EXIF so no chance for mistakes being left)