selmslie Loc: Fernandina Beach, FL, USA

The presentation made that abundantly clear.

The problem is that Ysarex is zeroing in on a single point - that the exposure (and ISO) used by the camera are different from what the "Exposure" slider does in post processing. DUH?

Of course they are different! Their used in a different context. One applies to taking the image and the other to post processing,

His beef is with the people at Adobe who used the term "Exposure" for the slider. All other post processing programs followed suit.

He is spitting into the wind, beating a dead horse.

TriX Loc: Raleigh, NC

Scotty, I’ve watched the video twice, and I don’t think it’s “abundantly clear”. It may be a “duh” to you, me and Yarex, but it’s NOT abundantly clear to many members as evidenced by past posts on ISO invariance, hence the reason for my post - not to denigrate the video, although some of its simplifications and statements are not actually correct, but just to emphasize that the two methods are NOT fungible and do NOT have the same effect.

Ysarex Loc: St. Louis

In which case it's OK to present the facts wrong? I can appreciate that would be your position.

Ysarex Loc: St. Louis

I believe I identified a list of points.

selmslie Loc: Fernandina Beach, FL, USA

I only watched it once and it seemed like he clearly showed the differences between shooting at a higher ISO and using the same exposure at a lower ISO and using the Exposure slider to get them to the same brightness.

He also made it clear that there is a difference in noise between getting plenty of exposure at a lower ISO and less exposure at a higher ISO.

And his explanation of the doglegs in the DR vs. ISO line was also pretty clear in that he pointed out that there are different algorithms being used to do the A/D conversion that can adversely affect the noise reduction if you try to beat the system. He didn't user the word "algorithm" because he would have had to digress and explain what that meant.

So he is not being misleading or ignorant He covered most of the stuff pretty well.

He was targeting a general audience rather than a group of technical people who already understand how this works. He did his best to cover the topic with some occasionally quirky metaphors.

We should cut him some slack.

You or Joe should try and put together a presentation of your own that is absolutely flawless that the general public can understand.

selmslie Loc: Fernandina Beach, FL, USA

And I'm not sure that your list is accurate but I'm not obsessed with proving that either of you are wrong so I am not going to waste my time watching the whole thing again taking notes. That would be silly and obsessive.

I don't think you can do any better considering the audience he was targeting. But if really you think you can, prove it.

TriX Loc: Raleigh, NC

Scotty, i have/had no intention in participating in another of your endless arguments - I was simply emphasizing the difference, so I won’t be responding further. This will go on forever - if Yarex said the sky was blue, you’d initiate a 10 page thread arguing it was red. Your endless arguments obscure the useful in depth information you periodically post.

In fact, if you want to get into the weeds, your use of the word algorithm to describe the difference in the low and high gain amplifiers is plain wrong - they are separate HW devices (although not in the way described in the video). If you want to understand how a dual gain sensor actually works, you should read the links (and listen to the designer’s audio explanation) that Yarex posted. And since we’re discussing minutia, it would have been just as easy for the presenter to have actually described accurately the way the system really works.

You’re very fond of technical minutia and posting detailed technical analysis in detail, so I’m frankly surprised you’re going out of your way to defend some inaccurate generalizations under the heading that it’s simplification for the masses. I wonder if you’d be so vociferous if Yarex hadn’t brought it up.

I’m done - goodbye.

Racmanaz Loc: Sunny Tucson!

Did you even watch and pay attention to the entire video? He does explain all this in detail

Ysarex Loc: St. Louis

What you say here is false. He did not use the word algorithm because he specifically talked about hardware that he made up that doesn't exist. He begins about 3 minutes and 25 seconds into the video where he presents not Photons to Photos DR vs. ISO chart for the XT-3 but the Read Noise chart. He points out the dogleg at the ISO 640 switch and notes: "There's less noise. And then the noise continues back up. What is happening here? Why does it drop like that. That is where a different amplifier kicks in. Your Fujifilm camera has two different amplifiers."

While saying the above he displays his hand draw diagram that shows two separate analog signal amplifiers prior to A/D conversion. He continues: "And that other one that kicks in, there's a little bit of a drop. It actually gets better quality at first before it goes up. If you're shooting an ISO less than 640, it's going through this amplifier. [points at diagram] If you're shooting an ISO greater than or equal to 640, you're going through this amplifier. [points at diagram] So the signal gets amplified and made even brighter. Then it goes to your analog to digital converter."

He gets very specific about hardware that doesn't exist in a Fuji XT-3. The camera does not have two separate analog signal amplifiers to implement ISO. (I provided a prior link that explains how the sensor functions.) He continues: "So I had to explain all this to you because cameras that are ISO invariant have very little noise usually around this step and this step in the process." [points at diagram section after analog signal amplifier and after ADC]. He's being hardware specific here and he's wrong. The reason some cameras are approaching ISO invariance is because they have less read noise before the signal is passed to the analog amplifier. In the meantime he managed to slip in that the signal picks up photon noise as it moves through the camera.

Yes, he said he's oversimplifying. But what he's really doing is misinforming because he doesn't understand.

P.S. I notice TriX called you on this one as well and he sees through you. https://www.uglyhedgehog.com/t-712451-2.html#12577350

selmslie Loc: Fernandina Beach, FL, USA

He is just using a metaphor to explain the dogleg(s). There are neither two amplifiers nor two read channels. There are just one or more algorithms to cover the normal ISO range. But neither of you are familiar with the software.

The Fuji plots show evidence of 18 cameras that have single gain sensors, 13 that are dual gain and one that appears to be triple gain based on the plot (see above). That doesn't mean that there are multiple channels but rather multiple algorithms used within the normal ISO range. So when he refers to two "amplifiers" he is just referring to two normal ISO ranges that behave differently.

But even Fuji cameras with plots that don't have a dogleg are already using multiple algorithms to provide JPEG results beyond the normal ISO range. About half of the them do this to extend the effect of ISO and exposure on the brightness of the JPEG beyond the normal ISO range.

He is also correct in saying that there is, "very little noise usually around this step and this step in the process." The highest Fuji ISO where the dogleg drops is at ISO 800. There is nothing wrong in stating that there is very little noise at around ISO 640 or 800.

If you think you understand it better than he does, why don't you try explaining this phenomenon. Can you do it better than he did without metaphors?

selmslie Loc: Fernandina Beach, FL, USA

PS: If you want to see multiple doglegs, take a look at the Nikon D3, D3S, D5, D6 and D700.

Ysarex Loc: St. Louis

You are wrong. It is not done using software algorithms. It is done in hardware on the sensor. The charge in each pixel can be read at one of two different capacitance levels. (sounds like channels to me). In the case of the Fuji XT-3 the capacitance switch (hardware not software) occurs at ISO 640. Here's some reading: https://www.sansmirror.com/newsviews/2016-newsview/jan-mar-2016-newsviews-2/dual-gain-becoming-the-norm.html

selmslie Loc: Fernandina Beach, FL, USA

Did you even look at that five-year-old link? It's full of errors and misstatements like, "The Fujifilm bumps gain at ISO 800, the Sony at ISO 400."

Fuji actually bumps the gain at ISO 200 (1 camera), 250 (1), 500(3) 800(8) and at both 320 and 1000 (one camera). Apparently neither of you looked at the Photons to Photos information.

It's like the blind leading the blind. You don't understand the subject any better than he did five years ago.

But the bottom line is that this is the way that the camera deals with DR and noise. It has nothing to do with ISO invariance or the camera's linear response to ISO. All of this fog and darkness you are throwing at the question of Fuji ISO Performance is irrelevant.

I have yet to find a camera that violates the principles of the exposure/ISO relationship for the raw file within its normal range.

Ysarex Loc: St. Louis

Did you maybe read the first sentence: "Both the recent Fujifilm X-Pro2 and Sony A6300 demonstrate something that’s becoming more common in sensors: dual digital gain." He's talking about two specific cameras.

Here's the Photons to Photos link where you can see the difference between the Fuji XPro-2 and Sony A6300 as he correctly noted: https://www.photonstophotos.net/Charts/RN_ADU.htm#FujiFilm%20X-Pro2_14,Sony%20ILCE-6300_14

Both cameras use a Sony sensor that implements Aptina's DR-Pix dual capacitance switching which explains the dogleg in the graph that the video author was commenting upon. It is implemented in hardware on the sensor. It is not a software algorithm and you are wrong.

selmslie Loc: Fernandina Beach, FL, USA

I don't have to read someone else's interpretation of what they think is happening during a digital capture. As an engineer, I understand exactly what is involved.

We know that when a sensor receives 500, 1000, 2000 and 4000 photons at a the sensor (based on precise settings of exposure and luminance) those should be separated by exactly one stop in the raw file within the camera's normal range.

But those photons produce a charge on the individual sensels that may not be separated by a charge that is perfectly in proportional to the number of photons.

The job of the A/D converter is to compensate for any difference in the charge, after reading from the sensor, as being in exact proportion to the number of pixels received so that the raw data is recorded in a proportion of 1:2:4:8.

Whatever algorithm can successfully do this at a high ISO setting might not work as well as at a low ISO setting. In order to accommodate this two or more algorithms might be necessary to correct this and provide a linear response that keeps the ISO/exposure relationship linear over the camera's normal range.

Imagine how difficult this was to accomplish for the Nikon D6:



Nikon took the trouble to get this done for one of their flagship digital cameras because they felt it was important to get it right. It took several distinct algorithms to make it all work. It ended up with six doglegs in the plot. That's not just a dual gain sensor.

The engineering marvel is that this could be accomplished for all of the 20+ megapixels that the camera records over a normal range from ISO 100 to 102400 and an extended range far beyond those limits.