Photography Chat

« first day (2094 days earlier) ← previous day next day → last day (2866 days later) »

dpollitt

2:10 AM

Something going on with the alignment of the header image and the image text.

8 hours later…

Euri Pinhollow

9:53 AM

http://photo.stackexchange.com/questions/77765/canon-aps-c-maximum-sensor-size/77767?noredirect=1#comment133675_77767

If one calls any process of getting full-colour image from Bayer sensor demosaicing, it sure is required to evaluate colour sensitivity.

However, evaluating colour sensitivity does not require getting RGB image at all, it only requires gathering statistics from the channels and computing resulting values.
>Just go to DxO Mark and read their own methodology descriptions.

I have once saved all of their method descriptions. I have searched for words "[de]bayer[isation]" and …

>If one calls any process of getting full-colour image from Bayer sensor demosaicing, it sure is required to evaluate colour sensitivity.

Scratch that. It is not required, I cannot remember why I wrote it.

Michael Clark

@EuriPinhollow They claim they don't do any demosaicing, yet they also claim to measure color sensitivity and color response. What's wrong with that picture? They also weight 2/3 of a sensor's overall score with things only measured at ISO 100 (Color Depth and Dynamic Range).

Even the "Low Light ISO score" is measured by using an arbitrary set of conditions: "Thus low-light ISO is the highest ISO setting for a camera that allows it to achieve an SNR of 30dB while keeping a good dynamic range of 9 EVs and a color depth of 18bits."

Since no one uses raw monochromatic values from a Bayer masked sensor without demosaicing first (even when the final image is a monochrome one), what DxO Mark claims to measure without doing demosaicing is meaningless.

Further, normalizing everything to 8MP for the "print" measurements means any advantage due to resolution, either from sensor resolution or lens resolution, is totally negated.

Euri Pinhollow

10:47 AM

"a sensor's overall score" - derivative, unrelated.
"Low Light ISO score"" - derivative, unrelated.
"Since no one uses raw monochromatic values from a Bayer masked sensor without demosaicing first" - I cannot see any ground for it. You are extrapolating practical photography onto studying the technology. You really need to tell me why exactly everyone let DxO alone should use debayerisation for that. Remeber, DxO guys do not even use colour profiles for dynamic range measurement (and noise depends on the profiling as you know). They only use applied profile for "Full CS [Colour Sensitivity]…

5 hours later…

Shyju

3:37 PM

Ahoy

Getting my new tripod today

;)

Can't resist to buy the NDFilter set !!!

Lee's set is so expensive :(

Any other suggestions ?

2 hours later…

Michael Clark

5:20 PM

@EuriPinhollow dxomark.com/About/Sensor-scores/Viewing-Conditions

@EuriPinhollow dxomark.com/Reviews/…

Euri Pinhollow

@MichaelClark does not change the fact that they do not resize image to get normalised SNR. Please read the page which I mentioned.

@MichaelClark oh well, you linked the same page which I did.

Michael Clark

They effectively do. If you understood the equations on that page you would see that what they do is.

Euri Pinhollow

Yes I do understand what equations mean.

I also understand that no equation means resizing (with unknown algorithm).

@MichaelClark "However, any other resolution can be chosen, as doing so only shifts the normalized values by a constant (because the reference resolution appears only as a logarithm in the formulas above). "

Michael Clark

@EuriPinhollow "...we chose a reference resolution equal to 8 Megapixels, which is a bit less than a 12" x 8" print with a 300dpi printer. "

Euri Pinhollow

Point stands.

Michael Clark

5:26 PM

No, the point is the numbers published by DxO for "Print" measurements are based on normalization to 8MP.

Euri Pinhollow

So what?

If they were normalised against different resolution, what would change?

Michael Clark

The numbers. The difference between 48dB and 42dB sounds a lot less significant than the difference between 24dB and 18dB, even though both comparisons are one stop.

Euri Pinhollow

DeciBells s are additive, not proportional. Whoever pays attention to ratio between two values is uninformed.

Michael Clark

5:42 PM

Whoever thinks that differences in dynamic range and color sensitivity at ISO 100 carries throughout the entire use range of different cameras and should comprise 2/3 of a camera's sensor score are also woefully uninformed. But there seem to be a lot of folks out there that buy a camera based strictly on the DxO Mark sensor score.

AJ Henderson

@EuriPinhollow not sure what you mean, the point is that 6dBs is a doubling regardless of how many dB you have yet, but to an average person, they are going to think that it's a bigger change if the starting number is lower. People LOVE to oversimplify.

Michael Clark

Even when they are planning to use it in less than optimal light that will require ISO speeds several stops removed from ISO 100.

Euri Pinhollow

@MichaelClark I cannot help them. Sensor score is a derivative from known values, why not use them instead?

@AJHenderson Michael Clark says that DxOMark somewhat handicaps higher resolution cameras because of 8MP reference. This is not true for mentioned reasons.

Euri Pinhollow

5:57 PM

photo.stackexchange.com/q/77784/49477 - tired of people searching for problems preventing them to do what they like. storage9.static.itmages.ru/i/16/0526/…

Why can't I flag question as belonging anywhere else other than to Meta.Photo.SE?

2 hours later…

Euri Pinhollow

7:40 PM

photo.stackexchange.com/questions/29540 - I suggested an edit removing "anti-aliasing" from question for reason said in comments. It got rejected. Using "anti-aliasing" is misleading and should be removed from question.

Euri Pinhollow

7:55 PM

@AJHenderson I'd like to hear your opinion.

Michael Clark

@EuriPinhollow It may not be the exact way stack exchange is designed to operate, but when the user whose answer you are attempting to edit is the highest reputation user on the entire site, it's probably better to suggest the edits in a comment. Further, the inline edits make no material change to the answer. The note at the end could just as easily have been made as a comment.

Euri Pinhollow

@MichaelClark " it's probably better to suggest the edits in a comment." - welp, I did. There is no aliasing happening in this case. Would you fix "Canon D810" if you happened to see it? It probably would not make material change of many questions too.

Michael Clark

Not to mention that the answer was written over three years ago. The 16-bit option you mention may or may not have been available at that time.

Euri Pinhollow

@MichaelClark Excuse me please but do you actually read the Q and A which I talk about? "16 bit" was already mentioned at the time I have seen the Matt's answer. Photoshop CS5 is quite some years old already.

Michael Clark

8:13 PM

There's no need to be rude. Yes I read the question. Yes I read the answer, including your suggested edits. No I did not go and look up the release date of CS5. The point is that old answers are often written before information included in a suggested answer was available.

Whether or not that was the specific case here isn't very important.

*suggested edit

Euri Pinhollow

Point of my complaint about Matt's answer is that tonal smoothing is in no way guaranteed after combination of upscale/downscale is applied and in case of Photoshop it surely does not work.

Michael Clark

@EuriPinhollow There is a kind of aliasing going on here. It's just not with regard to the shape of lines but rather in the luminance values of pixels in an image. Aliasing is not restricted only to graphics. It is also applicable to signal processing. Audio processing, for example, can produce aliasing that results in harmonic frequencies not present in the original signal being present in the processed signal.

Euri Pinhollow

@MichaelClark I surely know what aliasing means. What kind of aliasing is going there? Which series of inputs yield only one output?

Michael Clark

@EuriPinhollow In which case the way this works is you are allowed to downvote the answer. You are not allowed to edit every answer on the site to what you feel is "correct". That's not how this works.

@EuriPinhollow tonal values.

Euri Pinhollow

Not talking about black and white points because it is desired aliasing.

@MichaelClark which series of input tonal values has single output tonal value?

@MichaelClark not only do I think that something is more correct than something else, I can answer any question about my point and can substantiate it.

Michael Clark

8:30 PM

@EuriPinhollow ranges of tonal values are being represented in 8-bit color space by the same value. i.e. any value between 16319 and 16383 in 14-bit space are represented as 255 in 8-bit space. When the middle values (i.e. the 20% through 80% values) are stretched to the full range of 256 values the finer gradations are not restored. Everything that was represented as, for example 210, is re-assigned the same new value such as 250.

None of the former pixels with a value of 210 are remapped as 249 or 251.

@EuriPinhollow Then write your own site. That's not how this one works.

Euri Pinhollow

I will reply in next few days, short on time ATM, I must rush for other business imidiately.

« first day (2094 days earlier) ← previous day next day → last day (2866 days later) »

all rooms