-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
The 10mp sensors are lifeless
- Thread starter aarif
- Start date
I think its too easy to want to spend far too much money above actual photographic requirements and that there are perfectly capable cameras and lenses available for far less than the 'latest and greatest', like the Canon L's and Full Frame models.
Its easy to get caught up in measuring gear and feeling left behind when new models with larger resolution and sharpness come out, or to look for marginal improvements which then become must have buys - this for me would be a huge drain of resources that could be better spent elsewhere.
After starting out looking for a dslr camera and lenses by being interested in resolution, megapixels, etc and buying a Canon L zoom and primes I've found that its whether the camera takes the kind of pictures I like that counts for me.
The 6 megapixel cameras are amongst the best in this regard so I'm getting rid of the Canon gear ordering a Pentax K100D, a good prime and two average zooms, I'm better off financially and can spend some time looking at excellent images for inspiration rather than looking for the latest specifications in the gear itself.
I'd recommend looking online for the kind of pictures you'd like to take yourself, then seeing what gear took them and buying the cheapest that does the job.
Its easy to get caught up in measuring gear and feeling left behind when new models with larger resolution and sharpness come out, or to look for marginal improvements which then become must have buys - this for me would be a huge drain of resources that could be better spent elsewhere.
After starting out looking for a dslr camera and lenses by being interested in resolution, megapixels, etc and buying a Canon L zoom and primes I've found that its whether the camera takes the kind of pictures I like that counts for me.
The 6 megapixel cameras are amongst the best in this regard so I'm getting rid of the Canon gear ordering a Pentax K100D, a good prime and two average zooms, I'm better off financially and can spend some time looking at excellent images for inspiration rather than looking for the latest specifications in the gear itself.
I'd recommend looking online for the kind of pictures you'd like to take yourself, then seeing what gear took them and buying the cheapest that does the job.
gheil
Senior Member
As far as i can determine he is not interested in any such approach. i think most of Phils bread must be buttered by jpeg people. i have suggested to him (replying directly to his posts) these same basic ideas, and never gotten a reply. There are a few magazines that do do more serious tests, testing basic sensing capabilities of cameras. But i think if you want to be really popular, as DPReview is succeeding at, you have to appeal to the jpg masses. Personally i do not see that such obviates a more serious thoughtful look at actual performance. He just does not seem interested. Not enough butter? Hard to get an industry to swing if you cannot even get the basic tests published in highly visible places. i just cannot understand what Phil has to gain by ignoring such basic, fundamental, comparisons. Technically involved, yes, but fundamental and not that much more so than his other tests.
If you want to see a programs that does conversions using a lot of different Bayer methods then look at Dalibors converter for Minolta raws:
http://dalibor.cz/minolta/index.htm
--
http://public.xdi.org/=greg.heil
gheil
Senior Member
Then his tailor would thrice clothe him fakely
And he would feel all the more foolish when exposed. i do think you can pass a threshold and never be able to reverse oneself. The mental pain would be worse than the cure. GWB will never understand the pain he causes, because he has to torture his mind so much to convince himself of his own righteousness. And further to be the mouthpiece for such pain and suffering. The admission is just too humiliating. Nixon seemed to come around years after he was impeached, but these are extreme efforts. And only done obliquely. GWB does not have that fortitude. Sony is a pragmatic company - when the bottom line shifts they will follow the trend, but not until it is profitable, in the overall picture. Opinion leaders have to do their job, and lead. Then the industry eventually have to follow. As it is they can happily make more money by just printing more densely, without ever selling the more complicated, nuanced, story of sensor responsiveness. --
http://public.xdi.org/=greg.heil
Grizzlybob
Well-known member
You mention having both a 6MP and a 10Mp camera, but contrast the 10D and 20D. They are 6 and 8 MP cameras, respectively. What 10MP camera are you comparing to?
Grizzlybob
Well-known member
Is it the Sony A100 (10MP) that you're comparing images with or another 10MP camera? I haven't seen any comparisons done directly of 6 vs. 10 on the image, but I'm sure there could be some negative consequences in the way different cameras set up for sharpness, saturation, etc. I've heard this more in relation to the image variables that can be set, not the overall image where general conclusions could be made. It's an interesting premise and, having not read all the responses, I don't know whether anyone has tested this. Then again, it may be related to the manufacturer's learning curve on image production from the cameras. I know my 20D seems a step up as far as pleasing images compared to my 10D. Thanks.
gheil
Senior Member
If one of the cameras was a 5D, the test is totally invalid. Like mixing apples and oranges to put a large sensor camera in the mix. The whole idea would be to keep the sensor size constant and vary the "sensel" size. Using a 5D would vary the heck out of it and i am sure Petteri would not make so gross a mistake. A Arif after all is a very sophisticated photographer who is using a Canon 5D, as well as a Minolta 7D and Sony A100 and the other 10MP cameras at least by reputation, and knows full well that its pixels are quite large and of a totally different class than the yucky 10MP sensors this thread is all about. He probably knows Arifs reputation and would not insult him with such a ringer.
--
http://public.xdi.org/=greg.heil
GordonBGood
Veteran Member
As follows and interspersed:
I don't know how you can say that Phil Askey is not interested in evolving his reviews with digital camera developments: he has continuously improved his testing and techniques of determining DR of DSLR's, Simon's reviews now have some analysis of preservation of low contrast detail by providing hair 100% zoom crops at various ISO's, he now provides a detailed image along with the grey patches in his noise analysis pages so one can see the detail preservation and sharpness as compared to various equivalent cameras, etc.
I believe that Phil regularily does a search for his name and that of DPReview to see what we are saying about him and this site, and may even now be consulting with someone with a mathematics background to see what can be done along the lines I have suggested (provided he has time after working on the backlog of DSLR camera reviews and the various Photokina reports). He may even contact me, and I would be happy to provide him with source of my work so far or my completed programs when I get time to finish them.
Regards, GordonBGood
Since one generally has control over how much NR is applied when given raw image files, the tool generally applies to JPEG's, but can also be applied to the output from raw convertors to determine how much detail is lost vs. how much the standard deviation of noise in flat textured patches is reduced.
I don't know how you can say that Phil Askey is not interested in evolving his reviews with digital camera developments: he has continuously improved his testing and techniques of determining DR of DSLR's, Simon's reviews now have some analysis of preservation of low contrast detail by providing hair 100% zoom crops at various ISO's, he now provides a detailed image along with the grey patches in his noise analysis pages so one can see the detail preservation and sharpness as compared to various equivalent cameras, etc.
I believe that Phil regularily does a search for his name and that of DPReview to see what we are saying about him and this site, and may even now be consulting with someone with a mathematics background to see what can be done along the lines I have suggested (provided he has time after working on the backlog of DSLR camera reviews and the various Photokina reports). He may even contact me, and I would be happy to provide him with source of my work so far or my completed programs when I get time to finish them.
I did look at Dalibor's work, and although impressive as far as user interface, etc., as well as no doubt being tuned for speed of conversion, he only uses various algorithms as commonly available in the public domain and doesn't really break any new ground.
Regards, GordonBGood
Possibly my thinking on this is too fuzzy to argue, but it strikes me that the method needs to be able to distinguish signal and noise of the overlapping spatial frequencies. Some of the more visually annoying image noise, especially after some NR processes, can be quite low contrast and more than a few pixels in extent. The same spatial frequencies that convey image sharpness. Perhaps I'm missing something in your method, in which case, sorry.
I take it that you normalise the results to the lowest ISO performance and work from there. I can imagine that is good with SLRs where the lowest ISO performance is already very good. I have a smaller sensor (2/3") camera that has enough noise at the lowest ISO to be a poor choice for normalisation, so I would not know how to compare those cameras.
But how do you account for anti-aliasing and the low pass function of the lens which will vary from model to model? That is why I wondered about using a known pattern, containing many spatial frequencies at a range of contrast to give the best chance of finding a method that can separate out MTF (not exposure dependent) and frequency-dependent noise (exposure dependent) at the same time - I assume that noise after NR is no longer white, so it is hard to separate them by any simple method.
Or is that even possible?
Another approach could be to calibrate the transfer functions of lens and anti-aliasing filter in a very low-noise situation (strong exposure of a low contrast target at lowest ISO) followed by noise measurement in a weaker exposure. Is that really your method?
Actually perhaps that is not so hard - two different exposures of a low contrast test chart could suffice - unless the NR were found to be adaptive! Perhaps you could try that with your DCT method, if you did not do that already
Certainly standard test charts do not seem to have enough pattern - very little of the image is of the same spatial frequency as the visually dominant noise. Your idea sounds better in that respect.
I'm not very convinced that someone would use any of these methods seriously, there may even be licensing problems for "commercial" use if (as I suspect is likely) none of this is original!
I have no plan to do any real work on this, but good luck to you!
Ken
magicone
Forum Enthusiast
Maybe, instead of making broad generalizations, you should dump your C* and get something better!
Forrest41926
Forum Pro
Why are they different crops? If you're trying to make the point that these things are indistinguishable at the pixel level, then showing us such varried scenes, well, hides all the differences in how things are rendered, at the pixel level.
My guess: #2 looks best, #3 gets the silver, and #1 the bronze prize.
My other guess: what's in focus? In the second photo, with the yellow leaves, if we're talking about resolution, you can see the most here, on the right side of the photo, in the branches; but is this superior contrast, or superior resolution? The first image, it looks like nothing in the crop is really in focus, and, well, it looks like part of the "acceptable" zone when you hyperfocus. I'm also saying this because it seems in this one like you were shooting at closer range, which means limited DOF. Same with the third image, to a lesser degree.
I'm assuming these are 100 % crops, so if this is 100,000 out of 6,000,000 pixels, you're still in good shape. Although the photo with the trees and yellow leaves is clearly better. It also looks more vibrant in terms of color, but of course that's the lighting and subject matter.
John Doyle
Leading Member
But only one manufacturer makes 6, 8 and 12MP DSLRs, and the only 12MP camera they make has a very big sensor. He could have been using a D2X I suppose.
DavidMillier
Forum Pro
http://www.slack.co.uk
His photographs have been taken with many different cameras including E10, D100, D1x, D2x, D200, D50, E1, E330, Kodak 14n and nx and various compact digicams.
The remarkable thing about his shots is that they all look very similar in presentation. His particular 'look' and workflow survives the transition from camera to camera. His 10MP D200 shots are no different.
This tells me that a great deal of supposed characteristics unique to a camera or sensor are NOT the fixed things some people suggest they are.
Galleries and website: http://www.whisperingcat.co.uk/mainindex.htm
His photographs have been taken with many different cameras including E10, D100, D1x, D2x, D200, D50, E1, E330, Kodak 14n and nx and various compact digicams.
The remarkable thing about his shots is that they all look very similar in presentation. His particular 'look' and workflow survives the transition from camera to camera. His 10MP D200 shots are no different.
This tells me that a great deal of supposed characteristics unique to a camera or sensor are NOT the fixed things some people suggest they are.
--
Galleries and website: http://www.whisperingcat.co.uk/mainindex.htm
charles johnson
Member
This is sounding all too familiar. In the near past this was stated as "Boy, those color negatives sure are lifeless compared to color slide film"
cdj
cdj
Gary Martin
Veteran Member
I agree. The so-called "lifelessness" of an image has a lot more to do with the shooter than the sensor.
--
http://www.pixelstatic.com
John Sheehy
Forum Pro
Post-processing and downsampling images to 800*600 hides most of the differences between cameras.
--
John
Gary Martin
Veteran Member
Well, the OP probably saw the so-called lifeless 10mp images on the web, which were no doubt post-processed and downsampled as well. I'm betting that Jon Slack's work looks great in print too, regardless of the camera and sensor used.
--
http://www.pixelstatic.com
Ed Leys20479
Senior Member
Gary,
You should read the OP's post again. It's not very long.
In it, aarif says "I have both a 6mp and a 10mp dslr and the 6mp wins easily for me".
But conjecture is fun, isn't it.
Ed
--
http://www.blackmallard.com/cal_ls/
California Light and Structure
http://www.blackmallard.com/o_barn/
One Barn
optical_illusion
Senior Member
Things might be improved in the future, but the OP was talking about the currently available 10mp dslrs.
GordonBGood
Veteran Member
As interspersed:
I find that when there is NR applied, the ratios between the high frequencies to the lower ones will change from these ideal ratios in ways that indicate that there is much less spatial resolution; Bayer interpolation that uses more smoothing is, in a sense, applying some NR due to its native processing.
Regards, GordonBGood
My technique currently works best when the image has noise, which provides a high frequency pattern which is not limited by AA filters or lens glass. I use as my standard a theoretical camera that has only gaussian noise, and treat the Bayer pattern with a minimally smoothing Bayer interpolation to obtain RGB values. I then pass that through the DCT conversion to get "standard" ratios between the highest "frequencies" and lower "frequencies".
I find that when there is NR applied, the ratios between the high frequencies to the lower ones will change from these ideal ratios in ways that indicate that there is much less spatial resolution; Bayer interpolation that uses more smoothing is, in a sense, applying some NR due to its native processing.
As you say, many new high MP compact cameras have a lot of noise, even at low ISO sensitivities, but that helps plot trends of reduced higher frequencies as sensitivity and NR goes up. I think that ultimately, I will need a very high frequency low contrast test target in order to try to determine how much low contrast detail is lost when noise isn't very detectable. For instance, in all of the 10 MP compact cameras for which I have seen reviews to date, I see a considerable loss of low contrast detail, even in the lowest ISO shots. Note that Simon now includes comparisons of hair shots at various ISO sensitivities for new small sensor compact cameras; I would like to see Phil do this or something similar for DSLR's.
But AA filters and the low pass function of lenses will affect high contrast detail, so be comparing high contrast detail to low contrast detail, one can see how much low contrast detail is being lost to NR. As to separating frequencies, the DCT responds to any loss of high frequency detail, so is a valid tool.
Yes, with a standard test chart, one would have to have both high contrast and low contrast patches in order to be able to plot the loss of low contrast detail as compared to the high contrast detail.Another approach could be to calibrate the transfer functions of
lens and anti-aliasing filter in a very low-noise situation (strong
exposure of a low contrast target at lowest ISO) followed by noise
measurement in a weaker exposure. Is that really your method?
Actually perhaps that is not so hard - two different exposures of a
low contrast test chart could suffice - unless the NR were found to
be adaptive! Perhaps you could try that with your DCT method, if
you did not do that already
Well, there shouldn't be licensing problems with the DCT, which is pretty old math. Any I don't plan to offer a commercial product to analyse the amount of NR; I just want to make a tool available that can help us access whether we are really making any gains in low contrast resolution with these new higher MP cameras.
Regards, GordonBGood
Keyboard shortcuts
- f
- Forum