|
Forum: Pentax K-1 & K-1 II
05-18-2018, 05:02 PM
|
| |
But this quarter is worse that the quarter of one year ago which seems to corroborate the CIPA figures of a mirrorless slump.
|
|
Forum: Pentax K-1 & K-1 II
05-18-2018, 04:44 PM
|
| |
And yet the most recent quarter was a downer so that "steady growth" ain't so steady.
|
|
Forum: Pentax K-1 & K-1 II
05-18-2018, 09:17 AM
|
| |
If the Pentax camera is only good for "stills and dead things," Then the Sony cameras are only good for indoor things. Their badly-engineered fragile bodies are not robust enough for outdoor use where it rains, snows, sleets, sprays sea water, etc.
And some of us users definitely do prefer IBIS for it's superior functionality (e.g., correcting roll movement of the camera) and amazing add-on features such horizon leveling, pixel shift, and astrotracer. Why add all the cost of stabilization to every lens with an inferior technology when a better technology paid for once can stabilize every lens? Canon's and Nikon's stabilized lenses are only good for making Canon and Nikon a lot of profits.
All of this is just like the silly mirrorless-is-small hype. Yes, the camera is smaller, but then almost every lens is bigger to compensate for the under-sized focal flange. Maybe it's OK for photographers with just one kit lens, but the more lenses you have, the more bulky a mirrorless system becomes because of all the added volume in every lens
P.S., Fuji's sales of digital cameras are dropping. Their financial reports prove people are not ordering them like crazy.
|
|
Forum: Pentax K-1 & K-1 II
05-17-2018, 11:22 AM
|
| |
Interesting images but there's another explanation for the missing dust.
The histogram for the 100 ISO image shows both a slightly boosted exposure and a shadow-enhancing tone curve that make the dust much more visible. The dust is there in the 800 ISO image.
|
|
Forum: Pentax K-1 & K-1 II
05-16-2018, 09:39 AM
|
| |
Perhaps I did not explain what I meant very well because what I meant does cover the low-SNR condition and certainly covers those interesting low-SNR images that you posted. Any analysis of pixel-to-pixel differences automatically handles low-SNR features such as those speckled radial triangles. That's what I meant about "structure."
Perhaps the strongest proof of the power of the accelerator chip is to look at a very low-SNR image taken with cameras without the accelerator (K-3 & K-3ii) versus with the accelerator (K-70 & KP).
Image comparison: Digital Photography Review
In the case of K-3 & K-3ii, most of the detail is lost in the noise and it's hard to make out the overall shapes of the grey logo on light-grey background. With the accelerator, the image has much better edge detail, much better shape preservation, and much lower noise.
P.S. It's also worth noting that the visibility of those low-SNR triangles says more about the human brain's ability to extrapolate expected detail than it does about images themselves. A low-SNR image that does not have a simple shape (e.g., that DPR logo) is much less easily recognized.
|
|
Forum: Pentax K-1 & K-1 II
05-13-2018, 06:34 AM
|
| |
It looks to me like the anti-aliasing simulator is turned-off or set to the low value in some of your photos. There's Bayer CFA color aliasing in a lot of the hair. (My apologies if I'm wrong).
As for the RAW developer, those seem to have evolved from simple interpolating demosaicers to much more complex algorithms that attempt to guesstimate color from the noisy, CFA-aliased data. The advanced RAW developers are trying to make pretty pictures. Who knows what effects these complex algorithms might have on images from a camera that has something like the accelerator chip.
Has anyone ever tried to hack the EXIF of a RAW file to change the ISO? If changing the EXIF ISO changes the developed TIFF, then it suggests that the RAW developer is using the expected noise level associated with the EXIF ISO value to adjust how it deals with the image. That is, the RAW developer may have some denoising effects. But if those denoising effects are not properly tuned to the reduced-noise properties of K-1ii images, they may chew into image detail.
|
|
Forum: Pentax K-1 & K-1 II
05-12-2018, 06:13 PM
|
| |
Yes, we can agree that:- there is proof of image processing on pure noise images
- it is "smoothing" the noise (but any effects on the signal are not well known at this time)
- it is contentious (but unclear those contentions are justified)
- DPReview has done a crap job on the K-1ii somewhere between gross negligence and outright maliciousness
- Ricoh would be wise to provide user-controls from the chip
We seem to disagree that: - anything done in-camera can be done in post processing
- signal can't be distinguished from added noise
- the chip is doing a "nearest neighbour"-type denoising approach
Actually, if you look carefully, the FFTs are not consistent with this at all. Energies for the nearest neighbors are actually preserved almost as if the chip is denoising and then resharpening (or never smoothing those ). (Unfortunately, bclaff's tendency to normalize all the FFTs makes it really hard to judge what's happening. )
That's a good question!
The answer is in analyzing all the differences in values of pixels in a neighborhood. Noise has independent differences that are statistically bounded to modest values. Image detail has structured differences (and similarities) and are potentially unbounded in value. Any large magnitude differentials above some statistical threshold are certainly "signal." Any small magnitude differentials that are structured are also probably "signal." It's the small magnitude, unstructured differentials that can be attenuated to leave only the structured ones and the large ones.
As much as I totally respect MJKoski's photographic skills and applaud his efforts to push cameras to their limits, I'm not confident of his conclusions because I fear that his choices camera settings and the RAW developer may be contributing to the problem.
If the image of the small dust spec spans multiple sensels, it will be easily distinguished from noise. And if the image of the small dust spec only falls on one sensel, then the RAW image will be wrong anyway due to Bayer filter sparsity. If the lens is out-resolving the sensor so that the image of a tiny dust speck only falls on a blue pixel, then the color information about the dust spec is irretrievably gone. We're back to the missing red berry problem. Using the K-1 or K-1ii with a sharp lens at a sharp f-stop WITHOUT the anti-aliasing filter is guaranteed to create bad RAW data.
I like this test! What's interesting is that the phenomena of Stochastic resonance - Wikipedia implies that noise will actually help occasionally reveal the smoothed low-amplitude signal in some images of the stack. The loss of low-amplitude detail will be a lot less that one might assume and the gains in noise reduction in the rest of the image will certainly be nice.
|
|
Forum: Pentax K-1 & K-1 II
05-11-2018, 07:49 AM
|
| |
That's a good point. Yet I can understand why Ricoh did what they did for the following reasons:
1. the use of NR on K-3 RAWs did not seem to prompt any outrage.
2. the use of the accelerator NR on K-70 RAWs did not seem to prompt any outrage.
3. the use of the accelerator NR on KP RAWs did not seem to prompt any outrage.
4. if accelerator NR could be turned-off, how much shall we wager that DPR would have turned it off to "prove" the K-1ii was no better than the K-1?
|
|
Forum: Pentax K-1 & K-1 II
05-11-2018, 07:04 AM
|
| |
There's zero evidence of "smoothing" in bclaff's analysis because bclaff's analysis does not look at an image with any signal in it. There's only evidence of noise reduction. There's an assumption that noise reduction requires smoothing but that is unproven and may be false.
Attenuation of higher spatial frequencies in a noise image says nothing about attenuation of higher spatial frequencies of a signal if the filter is a non-linear. A nonlinear filter can have strong attenuation of noise and near-zero attenuation of signal. The only signal that is in danger of attenuation from these kinds of filters is signal of an amplitude so low that it's indistinguishable from noise. But if the signal is indistinguishable from noise, it will also be lost in an image without NR anyway. And even weak signals may be preserved by a well-designed nonlinear filter because the spatial and chromatic structure of signal is different from the spatial and chromatic structure of noise.
|
|
Forum: Pentax K-1 & K-1 II
05-10-2018, 06:52 PM
|
| |
For linear filters (e.g., averaging adjacent pixels together or various convolution-based noise filters), the relationship between noise reduction and detail reduction is a mathematical fact. And if Ricoh is only using simple NR, then people have reason to be concerned about the accelerator.
Nonlinear filters are a differ beast entirely. There's a lot of really clever signal reconstruction methods that use what is known about the statistical properties of noise versus the statistical properties of signals to maximize noise removal while minimizing signal degradation. Maximum likelihood estimators, for example, can look at a set of pixels and slightly correct them to make them less noise-like and more signal-like. And we must admit that the splotchy saturated speckles of chroma noise, for example, are extremely unlike any image. Moreover, even if you take a picture of chroma noise splotches, a properly designed filter would notice that the image is "too splotchy" for just noise and not entirely remove the splotches that are actually in the scene.
It's going to take a lot more that an FFT of the noise to characterize what the chip is doing and whether it affects images in any detrimental way.
|
|
Forum: Pentax K-1 & K-1 II
05-10-2018, 02:51 PM
|
| |
It's only proof of smoothing under two conditions:
1) the scene is indistinguishable from noise.
2) the accelerator is a linear filter.
AFAIK, those FFT tests only looked at noise images. They never tested it on a signal so they can't say what it's going to do with a signal.
|
|
Forum: Pentax K-1 & K-1 II
05-10-2018, 11:16 AM
|
| |
Sometimes a low ISO image is simply impossible. The photographer has no tripod, they can't use a flash, the lighting is low, subject is moving, a small aperture is needed to get DoF.
It's night 50% of the time of this planet, it's cloudy much of the time, people are indoors most of the day, it's usually dim on the forest floor, most scenes contain stuff a range of distances, and lots of stuff moves. And then there's need for high DR which is just as much a problem at sunset as at moonrise.
Bottom line is that the majority of all the possible photographs one might take in this world require higher ISO (and high DR at all ISOs). For most of the history of photography, photographers could not take photographs except in ideal sunny f/16 conditions because film was notoriously insensitive to light (film wastes about 97% of the light). Early digital cameras were worse than film in this regard. It's only in the past few years that the equipment has improved.
But most photographers are still locked in a mindset driven by the inadequate equipment of the past. Most photographers think they don't care about 102,400 ISO because they never take 102,400 ISO pictures. But they never take 102,400 ISO pictures because they don't have equipment that can take 102,400 ISO pictures. They unwittingly lose most of the pictures they could take because they believe such pictures can't work. But that's changing.
I, for one, care about 102,400 ISO because I know how many more photos I could take, how much more freedom I'd have to pick narrower apertures or faster shutter speeds, and how much more DR I could get from all ISOs.
|
|
Forum: Pentax K-1 & K-1 II
05-09-2018, 05:17 PM
|
| |
How in the world is this "bad news"?
Yes, it shows there is noise reduction which is exactly what Ricoh promised. And it looks to offer 1 EV in lower noise. That's great news.
But it does not prove there is any detail reduction which is why people's panties are in a twist.
|
