[Lowell Goudge]

see my response here

https://www.pentaxforums.com/forums/general-photography/119989-we-better-off-...ml#post1246543

I would have to question the whole issue without seeing a lot more results

[urje]

With the D40x (10MP CCD chip) right at the bottom in the f/1.4 graph (-0.75EV), to me it seems this can't be just a CMOS vs CCD thing, so I don't get where that conclusion comes from.

To be honest, with the amazing higher iso performance of todays sensors, I think the idea that the t-stop gain diminishes with higher f-stops won't let me sleep any worse at night. But what I do find interesting is their (resonable but as yet unconfirmed) suspicion that depth of field increases due to this effect (and, e.g., OOF highlights are smaller). This is something that isn't compensated for with less noisy electronics. Ofcourse this would be difficult to measure, and the effect at f/1.2 is still less than the gain in F-stop over f/1.4 (it seems the effect is about 0.25EV worse, so you gain a quarter T-stop with the half F-stop between 1.4 and 1.2) so f/1.2 will still get you a smaller DOF than f/1.4, but possibly less than expected.

(and, after I wrote this, I read Lowells response above mine, and I got another lesson in reading before I write anything...)

[Lowell Goudge]

Actually I have the opportunity to disprove the whole thing with my kit


I have an *istD a K10d and a K7d. This means I can test 6 and 10 Mp CCD sensors that were used by pentax nikon and Sony as well as the 14 mP pentax sensor with my 50/1.4 k mont and my 50/1.4 super tak

Clearly pre digital lenses that the camera clearly can't know anything about and therefore won't make any adjustments for

[uccemebug]

Lowell, are you saying you think you can test for the bokeh issue and/or the light falloff issue?

[Lowell Goudge]

Lowell, are you saying you think you can test for the bokeh issue and/or the light falloff issue?

I can't say that I can test bokeh, any better than anyone else, but for light fall off, anyone with a K mount or M42 lens with F1.4 aperture can do this.

It is really simple. In fact, I do a similar test with my lenses now.

I shoot against a uniformly lit block wall, to test metering, and I have published the curve for my K50F1.4 lens many times on my *istD, K10D and K7D. All that is needed is to take the same setup and as opposed to checking exposure accuracy at each aperture, letting the camera pick the shutter the metering thinks it needs, take the lens at one metered point, (I would recommend between F4-5.6) and then change both aperture and shutter to match that exposure setting, across all apertures.

The problem is that sometimes F1.4 is just so bright that you don't have the range to take the shot because you are above 1/4000. but any way, you get the idea.

you can then measure in any good photo editor the greyscale value of the central portion of the frame. this may or may not change over the exposure range, and that change will be due to 1 of 3 specific things. Either shutter speed error, aperture error, or the behavior of the sensor as a function of the angle of incidence of the light. The other key indicator of sensor issues is not in the middle of the frame where the light hitting the sensor is highly accute, but at the corners. The issue of deep well sensors is largely a cause of vignetting. So a real indicator of this may be to measure not just the central portion of the frame, but the corners, and compare them. I do not knwo wat editor people use, but I use PSP X3 and it can measure either the entire scene or just the selection window, for greyscale values. Therefore when I do tests I use the selection window and select an approximate 10% area box in the middle. I have done checks when others have posted, on the corners also to show vignetting.

I am willing to bet that there will be much more issue with respect to lens accuracy, especially aperture, than with any other measureable issue. My gut feel is that Fast lenses might be a little "optimistic" in the nomenclature of the lens. Remember there is marketing involved here as well.

The fact is, every one should do a test with their kit to measure at least exposure accuracy with each lens/camera combo any way. Just to understand how it behaves.

Here is a copy of the Metering errors for my cameras with the K50F1.4 attached

You need to know that between roughly 25 and 230 greyscale value on the histogram, that the change in greyscale is almost linear in terms of EV with a 1EV change being about 45 greyscale. I measured this by changing shutter speed up and down with a fixed aperture and the same block wall target.

[Digitalis]

I am willing to bet that there will be much more issue with respect to lens accuracy, especially aperture, than with any other measureable issue. My gut feel is that Fast lenses might be a little "optimistic" in the nomenclature of the lens. Remember there is marketing involved here as well.

This in not unusual, the nomenclature surrounding f stops and the relation to the T stop, being defined as the absolute light transmission of the lens are often quite separate things. In order to do any genuine testing of any metering or sensor abnormalities knowing the qualities of the lens is paramount, because you are removing a significant variable. AFAIK Pentax calibrate the aperture mechanisms to be the most consistent at the widest aperture, f/4, f/8, and f/11. whether they are still doing this remains to be seen but from my tests the FA50mm f/1.4 is transmits approximately 94.0% giving a value of T=1.51

[Lowell Goudge]

This in not unusual, the nomenclature surrounding f stops and how they are calculated and the T stop or transmission of the lens are often quite separate things. In order to do any genuine testing of any metering or sensor abnormalities knowing the qualities of the lens is paramount, because you are removing a significant variable. AFAIK Pentax calibrate the aperture mechanisms to be the most consistent at the widest aperture, f/4, f/8, and f/11. whether they are still doing this remains to be seen but from my tests the FA50mm f/1.4 is transmits approximately 94.0% giving a value of T=1.51

If that is the case, then we should see a constant line at all apertures, assuming the lens is accurate.

remember the point here , at least in my test, is to see if there is a measureable difference in sensor perfromance at large apertures.

If We consider the hypothesis of the letter, that camera makers are boosting the ISO to compensate for the performance of the sensors for light at other than accute angles of incidence,

I suspect this will be most dominant at the corners, because that is where light will be at the highest angle off normal to the sensor, and would be vignetting.

If I use lenses that provide no aperture information to the camera, it does not know what is connected and therefore cannot compensate anything. Using the old pentax lenses, either K or M42 which are designed for film, should at least within the cropped area of the sensor be relitively good with respect to vignetting, so we have a "standard" we can test with.

Having 3 cameras 2 with CCD and one CMOS we can check first generation CCD against second generation, as well as CCD vs Later COMS.

Since we have the same lens, and the aperture is set by the lens, we can "normalize" the results because the lens is the same for all tests, and therefore errors in the lens should lead to identical results over all generations and technologies of sensors.

That's what I plan to test. For a control, I need to do this on a day where light does not change much, and I will measure the exposure with a reflected light meter, confirming the light does not change.

All I need is time.

[uccemebug]

Lowell, your planned tests sound like interesting stuff. Thank you for taking it on, I'm interested to see what comes of your work.

From the chart you've posted above, is there an "expected behaviour" from which the tested configurations are deviating? I would have expected that there would be a fairly linear correlation betwen "grey scale average exposure" and f-stop instead of the s-curve that has resulted.

[Lowell Goudge]

First test.

K50F1.4 on *istD, K10D and K7D

All cameras set to ISO 200 (Lowest on *istD) and best when playing with F1.4 lens.

The following was taken at 1/640 and F1.4 on each body

*istD Center 137 edge 101
K10D Center 121 edge 103
K7D Center 128 Edge 105

At F2 and 1/320

*istD Center 146 edge 137
K10D Center 132 edge 130
K7D Center 142 Edge 138

So the question I have is, is the low edge reading simply vignetting of the lens, as many lenses are prone to it wide open, or is it the sensor "deep well" theory, note also all bodies get very good just stopped down 1 stop (first click on the aperture ring)

I am inclined to think the differences are 1) vignetting wide open, and 2) errors in aperture.

3 cameras 3 sensor technologies, one lens. Note that 45 greyscale is 1 stop, the variances between all cameras either in the center or at the edge are awfully good considering all other errors. the one suprise is the K7 actually seems flatter, i.e. lower edge to center variance wide open, than the other two, which is reverse of the artical with respect to technology. Is it possible that it is really mostly a design of the micro lenses?

Edit note, I had made an error and re-entered the corrected K10D values

[DAZ]

It appears that everyone is doing these tests on/with Pentax cameras but the issue (from what the letter said) has more to do with camera from other manufacturers based on the data from DXO.
DAZ

[Lowell Goudge]

It appears that everyone is doing these tests on/with Pentax cameras but the issue (from what the letter said) has more to do with camera from other manufacturers based on the data from DXO.
DAZ

well, that's kind of obvious isn't it, this is the pentax forum after all.

Having said that, if it is a sensor issue and is on nikon and sony those cameras share sensor technology with the *istD and K10D, so it should also be on pentax by default shouldn't it.

Also, the paper states clearly the problem was with CMOS but the worst offender was a nikon CCD camera,

The point I was looking to quantify in my simple test was really did the problem exist or not. Having 1st and 2nd generation CCD and a later generation CMOS sensor all behave roughly the same, with one lens, and identical settings of shutter and aperture, to me makes me think this is if anything, a game played with exposure error and metering and not a sensor issue.

[uccemebug]

@Lowell, those are interesting numbers indeed. In one respect, the morale of the story doesn't change: wide open on a fast prime is for effect and not sharpness.

Very impressive as you say that it catches up by f/2 which is still ~1.6 stops faster than my favourite K28mm f/3.5.

Not sure what to make of the fact that your tests find the opposite of the article's prediction. As you say, you used a K series lens so the camera can't be arbitrarily modifying the ISO. And I would have expected the edges/corners to be much darker and that the centre would be less impacted. The unevenness looks to my uneducated eye more like the difference between an outgoing but mature technology and an incoming, newer technology.

@DAZ I think that that's exactly why this testing has value: Canon & Sony will have many people doing their tests.

[DAZ]

well, that's kind of obvious isn't it, this is the pentax forum after all.

@DAZ I think that that's exactly why this testing has value: Canon & Sony will have many people doing their tests.

As this is a Pentax forum I can see why the cameras being tested are all Pentax cameras. I can also see the value of retesting and verifying what others are testing. I just wish to point out the following. Testing just Pentax cameras is a data point of 1 if the issue is how the different makers deal with a problem not just where the problem (CMOS CCD) is. Also a negative test on just Pentax cameras doesn’t show that there is no issue.

DAZ

[Lowell Goudge]

@Lowell, those are interesting numbers indeed. In one respect, the morale of the story doesn't change: wide open on a fast prime is for effect and not sharpness.

Very impressive as you say that it catches up by f/2 which is still ~1.6 stops faster than my favourite K28mm f/3.5.

Not sure what to make of the fact that your tests find the opposite of the article's prediction. As you say, you used a K series lens so the camera can't be arbitrarily modifying the ISO. And I would have expected the edges/corners to be much darker and that the centre would be less impacted. The unevenness looks to my uneducated eye more like the difference between an outgoing but mature technology and an incoming, newer technology.

@DAZ I think that that's exactly why this testing has value: Canon & Sony will have many people doing their tests.

what is really interesting is that the cameras all behave the same way. At wide open there is a big error center to edge which I believe is vignetting, or perhaps the issue with non perpendicular light on the sensor. by F2 that is gone, the only real difference between the cameras from that point on, is that the *istD always has the center slightly brighter than the edges buy about greyscale 4-6 where the K10D and K7D have the edge mutch closer, to the center.

To me, that suggests that perhaps the *istD suffers more from the non perpendicular light, which would make sense, because it is a first generation sony chip, and since that time a lot of effort has been put into the microlenses over each cell.

[Ole]

Time to pull out a film camera and test your K lens for vignetting?