Iliah Borg

Lives in United States AK, United States
Has a website at http://www.libraw.org/
Joined on Nov 11, 2002

Comments

Total: 152, showing: 1 – 20
« First‹ Previous12345Next ›Last »
In reply to:

Tom Holly: All this for one stop? Life’s too short.

I shoot velvia occasionally. That has 5 stops of dynamic range. Hacking files to increase from 14 to 15 stops (or whatever it is). Couldn’t be bothered. I’d just buy a Sony if DR was that important.

@ Tom Holly : It's not "out of my way", it's called intellectual honesty. You have no supporting data, so you are trying evasion. Doesn't work. Nobody buys it. Bye-bye.

Link | Posted on Feb 14, 2018 at 04:55 UTC
In reply to:

Tom Holly: All this for one stop? Life’s too short.

I shoot velvia occasionally. That has 5 stops of dynamic range. Hacking files to increase from 14 to 15 stops (or whatever it is). Couldn’t be bothered. I’d just buy a Sony if DR was that important.

@ Tom Holly : what part of "spot-meter measurements from the scene and densitometry from the film" I asked for is not clear?

Link | Posted on Feb 14, 2018 at 04:39 UTC
In reply to:

Tom Holly: All this for one stop? Life’s too short.

I shoot velvia occasionally. That has 5 stops of dynamic range. Hacking files to increase from 14 to 15 stops (or whatever it is). Couldn’t be bothered. I’d just buy a Sony if DR was that important.

@ Tom Holly : Why should I take it anywhere, you said it right here, without any links. So it is up to you to provide supporting facts: spot-meter measurements from the scene and densitometry from the film. You can do it yourself, or provide a solid reference to this data. Kodak will be happy to amend their film data sheets I guess.

Link | Posted on Feb 14, 2018 at 03:19 UTC
In reply to:

Tom Holly: All this for one stop? Life’s too short.

I shoot velvia occasionally. That has 5 stops of dynamic range. Hacking files to increase from 14 to 15 stops (or whatever it is). Couldn’t be bothered. I’d just buy a Sony if DR was that important.

@ Tom Holly : Not according to my experience, and not according to Kodak's data sheets. If you have any measurements that prove otherwise, please show. If you don't, let's agree that Kodak know better what is that they cooked.

Link | Posted on Feb 14, 2018 at 02:08 UTC
In reply to:

Tom Holly: All this for one stop? Life’s too short.

I shoot velvia occasionally. That has 5 stops of dynamic range. Hacking files to increase from 14 to 15 stops (or whatever it is). Couldn’t be bothered. I’d just buy a Sony if DR was that important.

Portra has less than 11 stops, http://imaging.kodakalaris.com/sites/prod/files/files/resources/KODAK-PROFESSIONAL-PORTRA160nc160vc400nv400vc800.pdf

Link | Posted on Feb 13, 2018 at 23:06 UTC
In reply to:

falconeyes: I am sorry to disappoint, but the math is flawed.

> This means that, in essence, the camera is automatically capturing two shots, bracketed by about 1EV.

This is a misconception. The A, and A+B signals aren’t statistically independent. There is ZERO gain in DR over the A+B signal alone. Think of the two Raws A, B. The added Raw A+B does indeed have added DR over both A and B. But that’s the standard Raw to start with. There at best could be a minimum effect from reducing the read error if A and B are read additionally to A+B.

The entire thing is NOT like bracketing. It is zero extra info in for zero extra info out. The people who wrote the software unfortunately didn’t look at statistical independence.

Also, there is no reason to believe that A doesn’t clip where A+B does. Because A and B are formed from photosites of half size and half FWC wrt to a full A+B photosite aka pixel.

> there is no reason to believe that A doesn’t clip where A+B does

Well, facts are not a reason enough?

Link | Posted on Feb 13, 2018 at 02:49 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@ Dragonrider :

Do you think gain (V/e-) for A+B and A are different? One needs to feed something around 1V to ADC. If the gain for A is 2x as for A+B, signal level from A is as good as it gets. Could be the same for just B.

Binning does improve SNR in noisy conditions, that's true. But given the actual noise levels, IMHO it is not a very significant improvement.

Anyway, I'm pretty sure Canon are experimenting and will figure out the optimal mode. My concern here is more of providing support to whatever they use rather than giving them advice, I'm simply not qualified for that.

Link | Posted on Feb 13, 2018 at 00:55 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@ Entropy512 : you don't even need a high motion scene here, just a longish exposure I guess.

Link | Posted on Feb 12, 2018 at 23:51 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@ Entropy512 : right, truncation of A+B can occur in cases where neither A nor B are truncated. Green clips first in most of the daylight scenes.

Link | Posted on Feb 12, 2018 at 23:49 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@J A C S :
I feel uncomfortable when cameramakers are called stupid (and by the way when I see brand wars).

It is unfortunate, however, that neither Canon, nor others document their metadata. So unfortunate I may call it dull.

Yes, LibRaw, LLC (Alex and I) developed this app. Going by "RawDigger and FastRawViewer have released...", he is FastRawViewer and I'm RawDigger on even days, on odd days we switch places. Of course switching happens at 2400 GMT.

Link | Posted on Feb 12, 2018 at 23:35 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@Dragonrider : as far as I know, there are several things in development, including pixels with very small FWC, non-linear schemes, etc. In 2008 one of US innovators demo'ed a pixel with logarithmic response for general purpose sensors. I have 2 specialized sensors (I use them to shoot welding arc for research purposes) with photodiodes in photovoltaic mode, also log response.

Link | Posted on Feb 12, 2018 at 23:23 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@J A C S :
> http://www.fredmiranda.com/forum/topic/1453555/0
> Unless the information is not reliable

Before coming to conclusions one needs to thoroughly investigate not just raw data, but also metadata and DPP processing to make sure there is no compensation there for the said deviations. Interesting would be to analyze, say, tags 0x4015, 0x4016, 0x4027.

Link | Posted on Feb 12, 2018 at 23:06 UTC
In reply to:

steelhead3: Why only the 5dIV...many other cameras of Canon have dual pixels?

That's because other current Canon models with current Canon firmware do not record dual pixel raw files, so there is nothing we can extract.

Link | Posted on Feb 12, 2018 at 21:41 UTC
In reply to:

Slapstick Noir: More DR from 5DIV? That's an overkill...

Here is the "main" frame, magenta highlighting indicates the green channels of raw data being blown out, red highlighting is where both green and blue channels are clipped, black spots are indicating clipping in all the channels. From "Statistics" you can see how many pixels / % of each of the colour channels are clipped.

http://s3.amazonaws.com/IliahBorg/Screen%20shot%202018-02-12%20at%2012.36.01%20PM.png

Now, the "auxiliary" frame, clipping is very tolerable:

http://s3.amazonaws.com/IliahBorg/Screen%20shot%202018-02-12%20at%2012.36.22%20PM.png

Link | Posted on Feb 12, 2018 at 21:32 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@Dragonrider : Well, summing in a raw converter is also "after ADC", and allows to preserve the structure of the file. But we will see what will be the Canon's decision on this.

Link | Posted on Feb 12, 2018 at 21:24 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

Canon set black level to 512 for low ISO shots on this camera, for both main and auxiliary frames. The absolute maximum thus is 16383-512=15871. Thus the half is 15871/2+512=7935. I used an arbitrary number larger than that.

On a side note, the recommended normal white level (in Makernotes) is 13135, and the specular white is 14448 (both numbers include the offset of 512). This makes the choice of an exact number for the example rather difficult ;)

Link | Posted on Feb 12, 2018 at 21:18 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@Dragonrider : Yes, or they can sum after the ADC.

Link | Posted on Feb 12, 2018 at 21:04 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

@Entropy512 It is called "incremental camera upgrades". Wait for 5D Mark V ;)

Link | Posted on Feb 12, 2018 at 20:50 UTC
In reply to:

Sacher Khoudari: Doesn't make any sense to me. If the signal from A+B is saturated (full well capacity reached), then each A and B will be most likely saturated too. You won't gain any DR here. Nor on the lower end, because by doubling the number of pixels (A+B -> A and B), each pixel will have about 41% more noise.

If the signal from A is not saturated, but above raw data number 9000, and the same is for B, A+B is saturated, while A is not.

Link | Posted on Feb 12, 2018 at 20:40 UTC

Gentlemen:

The Studio Scene link for 7R3 pixel shift leads to DSC00234.ARW, which looks like an ordinary uncompressed 14-bit Bayer raw, no signs of pixel shift. Shouldn't there be three more raw files / data sets to form a pixel-shifted version?

Link | Posted on Nov 12, 2017 at 19:06 UTC as 184th comment | 3 replies
Total: 152, showing: 1 – 20
« First‹ Previous12345Next ›Last »