I've lost track of which is which. I emailed a list to somebody before I posted, so that I couldn't be accused of cooking the results. At some stage that person will jump out and reveal all, but hopefully not before a few more people have had a go.
I would, but the problem is the dressing room's so poorly lit, that I'd need a 1.5MP FF camera to cope, and I don't have one to hand.
I've just replaced my old CRT TV with a 37" 16:9 plasma screen (Panasonic X20). Its resolution is only 1024x720 (0.74mp, and the pixels aren't square!), and it's really quite amazing how good it looks with a good 720p/1080i signal, even at 3 feet distance. I'm pretty sure that a 37" (diagonal) 0.74mp print wouldn't look nearly as good at that distance, but not quite sure why that is. Probably something with the difference in contrast/DR between a print and a monitor, I guess.
I would call those controlled variables. In this instance, its not as though you are waiting until the experiment is over then saying "oops it never worked because we don't know the rate at which we heated the the temperature to 100C but if we had heated at 0.05C per second, we would have gotten the right answer. This is not about whether experiments have variables, this is about whether variables that affect the result are being controlled or not.
Thats not really how the gang comes across. People here informally share their own empirical evidence of the same few Fuji's they all use. If these people have found empirically that increased pd has reduced IQ, thats all they are sharing. They are not announcing a major experimental breakthrough.
What I refer to in Kims experiment is a series of posts from friends of the gang of 4 that listed a selection of variables that could affect IQ. This flagged the issue for me in a general way. The best one was possible differences in the substrata of the sensors.
Surely the doors, trim, fabric etc all have weight and are part of the car and that is what you are measuring, or have I misunderstood you? Are you just trying to measure the monocoque?
That is not the issue, what that sort of experiment does is allow you to make generalised statements rather than specific ones that only relate to a specific pair of cameras with a specific version of firmware.
Then they can explain to GB why the NR is an uncontrolled variable that can lead to contradictory results and conclusions for the same data set. They can also explain that if the variables between cameras are insignificant to the result, then they can't be used to explain away differences in the result whereas if the variables are significant to the result, then the result can't fairly be compared in the first place.
If this is an oblique reference to me for having the temerity to point out that because the experiment has not been peer reviewed and variables are not being controlled which means that wrong conclusions can be reached, all I can say is this is what can happen if you submit an experiment to a public forum, rather than to a respected scientific journal.
exactly, and to know that, you have to consider what is the likely impact of different variables on teh experiment on a case by case basis, and if they turn out potentially to be confounding variables, to control them. But isn't that exactly what the detailed discussion you've been having with Joe is about? You seem to be making a blanket charge that our demonstrations are subject to uncontrolled confounding variables. What you haven't done is enter into a discussion as to exactly which variable and why it is confounding in the context of that experiment . At the same time, you suggest that Kim's analysis of his expreiment has been unfairly dismissed on the basis of confounding variables, when the nature of the variables and how they impacted the result was very precisely presented.
I think you're generalising from your own perception. I've had enough communications since this discussion started to be reasonably satisfied that the 'gang's intervention has been seen as positive by some forum regulars.
I would differ from that view. People are presenting this 'empirical' knowledge as absolute and generalised fact, and it isn't. There were days when if the 'empirical evidence' 'proved' the ugly old woman to be a witch, she got burned. Thankfully, most places we've moved on from that attitude. The stakes might not be as high when it comes to cameras, but the faulty logic and confused motivation is the same.
Not sure what you mean. Performance parameters of sensors is a quite well understood thing, all that was happening was that well known and grounded knowledge was being deployed to show why the conclusion being drawn was false.
I'm not trying to do anything with cars at all. I'm trying to illustrate that when one wants to discuss an experiment, you have to be clear minded about what it is you're trying to prove or disprove before you even begin to discuss what the significant or insignificant variables are. 'We' really need to keep out eye on the ball, and the ball isn't about testing cars, it's about defining propositions before you do experiments to prove or disprove them.
It depends what is the generalised statement, as I said earlier.
There is no such thing as a general experiment, it is an impossible thing to mount. The way that science works is a body of theory is developed which fits together and provides a description of a phenomenon. Then, experiments are mounted to show that individual predictions of the theory are borne out. You will find no general proof of relativity or quantum mechanics.
They can't because it isn't. Which result which is being experimentally tested will NR invalidate and why?
No-one has said the variables between cameras are insignificant to the result. In fact Jop has told you several times that sensor efficiency is very significant to observed results.
You're arguing yourself round in circles. Variables are not universally significant or insignificant. Whether they are significant depends on what is the proposition being put forward . This discussion would fare better if rather than very vague and ill defined comments about significance of veriables you concentrated on specifics. Which proposition is under test, which is the confounding variable, why does it confound the test.
No.. I don't think you've been hysterical. I think you're been a bit woolly minded, and naive about the genre in which we're operating, but not hysterical.
That is a false accusation. As I said, be precise and prove your case before making accusations of peopel plying fast and loose. Which uncontrolled variables? How would the affect the claimed analysis? Put up or shut up.
Higher resolution? Oh ... you are now using resolution of the lens as opposed to pixel count and thus density.
So ... you have now given us yet another reason why larger pixels are superior.
Did you intend to add fuel to the fire? Or was that an accident ...
Hi Bob,
Hi Chris,
Yes, it is quite an honor.
This is a question that I've had for some time now, and I hope I ask ok: As technology improvements bring us to really low read noise levels, are there scale dependent (transistor sizing, etc.) as well as scale independent read noise components?
Yes, both the possibility of electrical and optical crosstalk are potential issues (please see the Lumerical website for cool wave simulations). But it's nice to see how brilliant innovation can help to overcome some of these things.
That would be cool!
Yes, this clears things up.
The point about read noise scaling is the mechanism through which it occurs. In terms of relative strength against other noise sources, it needs to be measured in electron equivalents, which means it get transformed through the input capacitance of the reading circuit, which is actually the base of a MOSFET and a little attached floating diffusion. Now, any read noise source downstream will effectively be transformed through that capacitance, so as long as the downstream sources are independent of pixel size then the read noise will scale down with the pixel (if the pixel is strictly scaled).
Apparently, not so much as I had hoped (further explanation follows).
A lower PD does not have less motion blur. For the same shutter speed, the motion blur is the same regardless of pixel count . However, unless the shutter speed scales with the (linear) pixel size, then the additional detail that smaller pixels can potentially offer is significantly reduced.
The higher PD often has an upside with no NR being applied, as it is more detailed, and a lot of photography is done at base ISO where detail matters significantly more than noise.
Now that is what I am saying! Well, almost -- NR is better than binning (in terms of IQ).
I think this has been already shown that higher PD results in higher resolution images given the same lens is used. See the dxomark and photozone lens tests. Same lens shows more resolution if combined with higher MP (and same sized) sensor in every single case.
http://forums.dpreview.com/forums/readflat.asp?forum=1036&message=37468394
"1/C = sqrt(1/L^2 + 1/S^2)
This uncertainty does not change the general conclusions reached by yourself, Anders and others.
There is no sharp cut-off when the sensor resolution exceeds that of the lens. Image resolution continues to improve as sensor resolution increases.
With the root sum of squares formula, we achieve 70% of the theoretical sensor resolution when the lens and sensor resolution are equal, 90% when the lens has twice this resolution, with diminishing returns above 95% when the lens has at three times the sensor resolution.
As technology improves it will be more cost effective to improve sensor performance than lens performance."
http://forums.dpreview.com/forums/read.asp?forum=1036&message=37448004
http://forums.dpreview.com/forums/read.asp?forum=1036&message=37448004
D700's resolution advantage over D5000 is due to sensor size. As D5000s image has to be magnified 1.5 times more (in linear dimension) than the D700's image to print at same size, the D5000 will always show 1.5 times less resolution given that same lens is used (and as D5000 has not more MP than D700). imaging -resource uses same Sigma lens on DX and FX cameras for their studio shots.
Higher resolution? Oh ... you are now using resolution of the lens as opposed to pixel count and thus density.
So ... you have now given us yet another reason why larger pixels are superior.
Did you intend to add fuel to the fire? Or was that an accident ...
Precisely so let me try to qualify better what I mean by uncontrolled variables so that we can eliminate Jupiter.
As I said before, I think you need to give specific examples of where a confounding variable has been ignored, and what effect it would have had on the result and analysis offered. Otherwise we're just dealing with generalised accusations of lack of objectivity and bias, and your 'experimental method' in coming to that conclusion is just as bad or worse that that which you are criticising.
I agree that lens performance does not seem to be the limiting factor ... you can tell the difference a stunning lens makes on the D70s as easily as you can on the D300 and the D7000 ....
1) D700 and D5000 create the same image of the same angle of view.
Remember, I have always given the condition of "same efficiency".
If an experiement confirms what I've said, then the differences in the variables is not significant enough to warrant further consideration. If an experiement contradicts what I've said, because it violates the "same efficiency" requirement, then a closer look at the variables is required.
I hope my points above have answered those points to your satisfaction.
