All true, but that is just explaining why there is a color shift, not why there is a different shift on the left and right side. The angles are identical - the only difference is if it is coming from the left side. The Bayer grid, as you confirmed is uniform so there could be no left/right side asymmetry.
The big question remains - why would there be a difference between the left and the right side of the sensor when everything is symmetrical and uniform?
There's also a top/bottom asymmetry
There is something called color correction matrix (CCM) which is applied to the RAW data - essentially position dependent adjustments for native pixel response. To oversimplify it, this is accounting / fixing for sensor vignetting. Lens vignetting is typically not wavelength dependent but sensor vignetting is dependent, and somehow this information has to be baked into the sensor response even before end user RAW data is supplied by the sensor. This data is measured and normalized for D65 illuminant conditions, and can sometimes be thrown haywire under a different color temperature. Not sure what your lighting conditions under which you imaged the flat field images in your earlier post, but sometimes these kind of artefacts seem to get amplified under different color temperatures, i.e. create artefacts like uneven color shading. I can talk more about this over private communication / send you technical literature if there is an interest.
It would be ideal to measure this response at D65 (approximately 6500K color temp.) and see what this shading looks like.
Thanks Ajay, that's very interesting and informative. However, I don't see how there could be a difference in wavelength depending on if a ray hits the right or the left side of the sensor..
The measurement by the way was done with 6500K through a diffuser to spread the light evenly.
A huge improvement on the C3 image while the M9 image is quite a bit worse than with ACR. RPP colors are terrible though - everything is puke colored with a bit of gray mixed in to liven things up.
Why doesn't Adobe or Sony or any of the other multi-billion dollar corporations buy up RPP. It's a amazing that an amateur project completely humiliates all the commercial stuff..
You may try picking a section to use the WB tool to try and get the colors a little closer. That being said, I don't believe that RPP has color transforms for the NEX-C3, yet, so the color probably isn't accurate for that camera, yet.
Adobe and others could probably make a similar engine, if they wanted to slow things down with floating point. Apparently, most of the big raw converter makers are sacrificing quality with speed.
p.s. Thanks, Luka, for taking the time to post all of these comparisons.
denoir wrote:
Why? The difference in resolution will be as with the other comparisons - both are high performance lenses that should easily outresolve both sensors.
should outresolve and do are different things aren't they thats why we do tests even if theory sais that it should be a certain way. i think the results will change, to where the nex will narrow the gap, but if you think otherwise i would to be proven wrong.
kahren wrote:
should outresolve and do are different things aren't they thats why we do tests even if theory sais that it should be a certain way. i think the results will change, to where the nex will narrow the gap, but if you think otherwise i would to be proven wrong.
i don't think the NEX can come close with an AA filter verses the m9 with none. i've never seen anything except the foveon sensors that come close to the m9 for per pixel sharpness. filtered sensors just don't look sharp at 100% the same way.
kahren wrote:
should outresolve and do are different things aren't they thats why we do tests even if theory sais that it should be a certain way. i think the results will change, to where the nex will narrow the gap, but if you think otherwise i would to be proven wrong.
There's not theoretical possibility that any lens will make a difference, as I've already shown the results with the ZM 25 - the world record holder for resolution (>400 lp/mm MTF10). It's diffraction limited at f/4 i.e it has reached the theoretical optical maximum. And as sebboh points out there is no way a filterless camera can be matched in per pixel sharpness to a filtered one. So unless I botched the test somehow the only way that there could be a difference is if I put a really good lens on the C3 and a really crappy one on the M9 (the 28 Cron does not fall into that category) or if I botch the second test.
Making a mistake during such a test is of course possible, but unlikely as I'm using a tripod + timer release in both cases and use 15x live view focus on the C3. Furthermore I'm using wide angles so missing focus at/near infinity is extremely hard.
M9, ACR. C3 IDC. I can't be bothered to do the C3 image in RPP as I then have to copy files to my Mac laptop, but you've seen the improvement that RPP brings in the test a couple of posts back.
douglasf13 wrote:
Raw Therapee is working on a new floating point version for windows that you could check out.
Last time I tried I couldn't get their 64bit version to start and was put off by all the bugs and frequent changes in the ways the different tools such as highlight recovery work. It seems to be a bit of a hack job of people who all have knowledge about some parts, but no one to make it into an integrated whole that doesn't crash all the time.
Regarding the difference in the sharpness of the 100% crops: if my 5D classic is a reasonable general measure for most APS-C or FF cameras that have an AA filter out there, _all_ of them will have a degree of softness similar to the NEX at 100% without sharpening. I never look at my 5D's files without some basic sharpening to recover what's lost by the AA filter.
It looks to me like no significant additional detail could be squeezed out of the M9's files by sharpening, but that's definitely not true for the NEX files, so why not show them with some capture sharpening which will invariably be applied by anyone processing the RAW files?
AhamB wrote:
Regarding the difference in the sharpness of the 100% crops: if my 5D classic is a reasonable general measure for most APS-C or FF cameras that have an AA filter out there, _all_ of them will have a degree of softness similar to the NEX at 100% without sharpening. I never look at my 5D's files without some basic sharpening to recover what's lost by the AA filter.
It looks to me like no significant additional detail could be squeezed out of the M9's files by sharpening, but that's definitely not true for the NEX files, so why not show them with some capture sharpening which will invariably be applied by anyone processing the RAW files?...Show more →
There has been sharpening applied in both cases - the default one that the RAW converter does. I think for tests like these if you start fiddling too much with the development parameters you risk making the test rather useless.
ACR's default sharpening for the M9 is suboptimal as well - you can increase the crispness quite a bit by using a smaller sharpening radius and increase the detail level.
The images were shot using long exposures from a tripod. Default settings for noise reduction in Sony IDC ("auto") were used. The images were then resized for web using my standard resize & sharpen script. Open the images in separate tabs in your browser and flip between them:
Not that brilliant, but it could be the IDC RAW converter. I'll wait for ACR support before passing judgement.
Did I mention how great the NEX is to use from a tripod? It's got the feel of a mini medium format camera when you flip up the screen. The M9 is awesome for handheld use but a major PITA for tripod based photography. In essence the opposite of the NEX - handheld live view focusing is as awful as tripod based rangefinder focusing. Now if Sony would come out with a FF version, preferably without an AA filter then we'd had the ultimate landscape photography camera (for tripod based shooting anyway). One can always dream...
denoir wrote:
There has been sharpening applied in both cases - the default one that the RAW converter does. I think for tests like these if you start fiddling too much with the development parameters you risk making the test rather useless.
ACR's default sharpening for the M9 is suboptimal as well - you can increase the crispness quite a bit by using a smaller sharpening radius and increase the detail level.
I see. Interesting, but then I must say that either the sharpening you applied to the NEX shots is very subtle, or if that is not the case the NEX-3C sensor creates disappointingly soft results (when comparing to my 5D).
AhamB wrote:
I see. Interesting, but then I must say that either the sharpening you applied to the NEX shots is very subtle, or if that is not the case the NEX-3C sensor creates disappointingly soft results (when comparing to my 5D).
It was the default sharpening of the RAW converter. I'm sure you could make it appear more crisp if you sharpened it more and/or differently. That goes for the M9 shot as well.
As for the per pixel sharpness it is slightly better than the 7D which is Canon's top of the line ASP-C camera. So I don't think it is soft when a reasonable comparison is made - a 5D (mk I) is a 12 megapixel full frame camera. It has fewer pixels distributed on a much larger area. It would have been astounding if it had not been considerably better per pixel.
