Page 3 of 4 FirstFirst 1234 LastLast
Results 21 to 30 of 34

Thread: ISO Invariance

  1. #21
    Super Moderator Kayaker72's Avatar
    Join Date
    Dec 2008
    Location
    New Hampshire, USA
    Posts
    5,565
    First off, understanding all this does help. Right now, I think the
    astrophotographers
    have it figured out, likely because it helps that type of photography the most. At least three astrophotography blogs I found focused on where a camera becomes ISO invariant. They have focused on the ISO where upstream noise (generated by increasing ISO) is greater than downstream noise. Looking at charts, I would agree with you/dpreview concluded, for the 5DIV, it is about ISO400.

    But, dpreview bases all their "ISO invariance" tests on ISO 100, ignoring the fact that different cameras become invariant at different ISOs. Maybe they get it and are only concerned about base ISO. Then, some of the tests TN/Fstoppers did could be explained by a dual gain sensor (TN figured it out in the second video).

    So, sensors have different characteristics, from megapixels, to pixel size, etc. I think this is exploring a lesser known characteristic.

    Quote Originally Posted by HDNitehawk View Post
    Maybe Sony is doing less control than Canon in body and that is why Sony is more "ISO Invariant".

    From the link in your last post:

    "How does the 5D Mark IV stack up against some of its other peers? Its performance is much better than any other offering from Canon, but it's still not quite as ISO-invariant as the Nikon D810 or D750,"

    Perhaps Sony has given the illusion that its ability to handle DR is greater than it actually is.


    This quote, it really is another version of the DR discussion just said in a different way with different terms.

    The 5D Mark IV isn't entirely ISO-invariant: pushing an ISO 200 underexposed by 5 stops by 5 EV in post-processing yields slightly higher noise levels than a native ISO 6400 exposure. An ISO 100 exposure pushed 6 stops fares even worse. However, above ISO 400, the camera does, for the most part, exhibit ISO invariance, meaning that you could underexpose a traditional ISO 6400 exposure by 4 EV by shooting it at ISO 400 (while maintaining the shutter speed and aperture for ISO 6400), and then raise exposure 4 EV in post. This technique would afford you 4 EV of highlight headroom, with little to no noise cost, relative to shooting at ISO 6400.


    I've wondered about Sony as well. The whole "star eater" issue they had, it seemed that their RAW algorithm was looking for noise and crushing it to blacks and happened to eat a few stars in the process. Of course, this is an artificial/destructive way to create black/shadows with less noise. From what I've seen, the current algorithm is better, but I do wonder if it is still going on, just more refined. Then, I am still looking at this "dual gain" system of theirs. If ISO is post pixel, it almost has to be a digital manipulation.

    I've also wondered about Canon vs Sony's Bayer filters. How many photons are lost at the Bayer filter? I do not even see this discussed (other than by those praising Foveon sensors that do not have a Bayer filter). But, thinking about it, theoretically if you defined "Red," "Green" and "Blue" as very narrow, your colors would be more accurate, but you would have to block out more light. Define Red/Green/Blue more broadly, more photons make it past the Bayer filter, but your colors are likely less accurate. So, in DXOmark, if you just look at the measurements for an individual camera sensor you can see their estimated accuracy in color reproduction (ISO 17321). For the cameras I've looked at, Sony and Canon are pretty equal, and no one talks about this, making me think camera manufacturers handle this similarly, likely with widely defined colors, but still something I've wondered about considering Canon is generally praised for colors.

    All that said, I do think Sony decreased downstream noise with on chip ADCs decreasing the travel distance for the analog signal. I've also read Sony has more ADCs than Canon typically has, which would decrease heat, noise, etc. Then, in looking through Thom Hogan's articles yesterday, I found this. If this ever comes to market, this next generation of sensor will read "0" as pure white and "max" signal as pure black. What this does is shifts low end noise to the whites, where it will hardly be noticeable and blacks will have almost imperceptible noise. This is cool. Technically, it is either an odd manipulate of the pixel wells, or they are doing something like reading pixels lost by the photodiode and not pixels gained and held in the pixel well. Something, but still, I have to give credit where it is do, this, at a minimum, is a cool concept.

    Quote Originally Posted by Karsaa View Post
    I took 2 pictures last night one with 4s iso 6400 f/1.8 and other same but iso 800 will post later on so that i sync wb and tint for both images and then boost exposure +3 on the iso 800 so we can have some comparision that lvl. body is 5dIV
    Great, I've been thinking of doing some tests, so it will be great to see yours.

    Ultimately, this was about understanding the system for me. It is interesting that several astrophotography blogs were definitely on top of this subject, but I do wonder how much of a difference it will make.
    Last edited by Kayaker72; 04-11-2019 at 01:10 PM.

  2. #22
    Senior Member
    Join Date
    Dec 2008
    Location
    Planet Earth
    Posts
    3,110
    Quote Originally Posted by Kayaker72 View Post
    First off, understanding all this does help. Right now, I think the
    astrophotographers
    have it figured out, likely because it helps that type of photography the most. At least three astrophotography blogs I found focused on where a camera becomes ISO invariant. They have focused on the ISO where upstream noise (generated by increasing ISO) is greater than downstream noise. Looking at charts, I would agree with you/dpreview concluded, for the 5DIV, it is about ISO400.

    But, dpreview bases all their "ISO invariance" tests on ISO 100, ignoring the fact that different cameras become invariant at different ISOs. Maybe they get it and are only concerned about base ISO. Then, some of the tests TN/Fstoppers did could be explained by a dual gain sensor (TN figured it out in the second video).

    Sounds familiar to the old DR discussions. Sony is obviously better.

    Most of the discussion is centered around camera bodies. Really that may only be half of the equation in the discussion. Don't you think it would be just as important to understand what is happening with your PP software? Or are we assuming all PP programs are equal and as such invariant? Do all the PP programs simulate the gain boost just like the camera?

  3. #23
    Senior Member
    Join Date
    Sep 2015
    Location
    Finland
    Posts
    437
    Here as promised:

    6400 4s f/1.8

    6400 by Kari, on Flickr

    800 4s f/1.8 +3 exposure

    800 by Kari, on Flickr

  4. #24
    Super Moderator Kayaker72's Avatar
    Join Date
    Dec 2008
    Location
    New Hampshire, USA
    Posts
    5,565
    Quote Originally Posted by HDNitehawk View Post
    [/LEFT]

    Sounds familiar to the old DR discussions. Sony is obviously better.

    Most of the discussion is centered around camera bodies. Really that may only be half of the equation in the discussion. Don't you think it would be just as important to understand what is happening with your PP software? Or are we assuming all PP programs are equal and as such invariant? Do all the PP programs simulate the gain boost just like the camera?
    Ha that sounds like another rabbit hole that I could spend awhile exploring. Which software is better. I will say, I did come across the following by Thom Hogan while reviewing the sensor performance:

    Here, he talks about the ISO invariance and brings up the software as well. Fairly interesting, his point (and the point of other program developers) is that programs like adobe operate as 16 bit even though our computers are typically 64 bit. This is to cut down on computational time and enhance the speed of processing. Granted, our files are 14 bit, but the articles point is that information is lost and working in higher bits is better for overall accuracy. They talk about RAWDigger and "Raw Photo Processer" being more accurate. Granted, he consulted at least one of the software engineers behind those processers for the article.

    Then Here, Thom states his opinion as to the best software varies by manufacturer, adobe caters to the largest (i.e. Canon), while Capture One has some deal going with Sony, etc.

    Quote Originally Posted by Karsaa View Post
    Here as promised:
    Thanks....the biggest difference is the purple streak going across the top of the scene in the ISO 800 shot +3stops. Contrail from a plane that went by?

    Otherwise, I still give it slightly to the native ISO 6400 shot. A bit more pattern noise while adding 3 stops to the ISO 800 image. I did have to look on flickr and zooming in helped, so close, but I'd give it to native ISO.

    Great colors by the way, the subtle transitions are excellent (oops, I know, I just talked about actual photography in a technical discussion...my bad....)

  5. #25
    Senior Member
    Join Date
    Sep 2015
    Location
    Finland
    Posts
    437
    Yeah, i can say that i prefer also shoot with iso i need not twidle with sliders =) I would say +3 gives more color noise. That is what comes when you boots exposure...or if not then my shutters is getting angry from last little time lapses. That line i mean.
    Last edited by Karsaa; 04-11-2019 at 05:37 PM.

  6. #26
    Senior Member
    Join Date
    Dec 2008
    Location
    Ottawa, ON
    Posts
    1,443
    Yeah, my first impression was more chroma noise too. I wondering, when you apply +3 exposure, and noise reduction, does ACR do the noise reduction before or after scaling everything * 8.

    What if you convert to 16-bit w/ just noise reduction, then once in Photoshop, boost the exposure. This will force the order of operations, and perhaps remove the increase in chroma noise.
    On Flickr - Namethatnobodyelsetook on Flickr
    R8 | R7 | 7DII | 10-18mm STM | 24-70mm f/4L | Sigma 35mm f/1.4 | 50mm f/1.8 | 85mm f/1.8 | 70-300mm f/4-5.6L | RF 100-500mm f/4-5-7.1L

  7. #27
    Super Moderator Kayaker72's Avatar
    Join Date
    Dec 2008
    Location
    New Hampshire, USA
    Posts
    5,565
    So, just ran some quick tests of my own. Now that I am inside and looking at the pictures, I am pretty sure my neighbor's motion lights came on for a couple of the images, but not others.

    That said....for the first time with my 5DIV, I am seeing banding. I had actually looked before. Not a big deal. Still love the camera and I really do not want to devolve to obsessing about pics raised 4-5 stops. While I was hoping there was some reason for the purple line in Kari's shot, it is likely banding as well.

    But, as I have been doing searches the last couple of days on this topic, I thought I'd do a quick one on banding. I found this, written by jrista, for those that were on CR a few years ago, you may recall, intelligent, detailed, DR focused posts that not everyone agreed with. Great astro images as well.

    As this thread is about understanding the process, if jrista is correct, it actually goes a long way to explaining things. According to jrista's response to a question, what we've been describing above, where ISO is gain added to the analog signal, well, that is Canon (and maybe other manufacturers). It is not the process for Sony.

    According to jrista's post, Sony did away with analog gain (ISO) with the exmor sensor. Not only did they increase the number of ADCs (one per column, which I've read before), they took the sensor output and directly digitized it before adding gain.

    So, of course Sony is "iso-invariant," ISO in exmor land is in camera digital gain vs post processing digital gain.

    But, they are also doing another "trick." And this gets back to the article Rick posted which I'll have to go back to read as I did not fully understand this part. But after a pixel is read, then there is a reset charge. Well, Sony is reading that reset charge and subtracting it from the next images signal. Thus, random noise generated while reading the reset charge (plus the reset charge) is subtracting out noise from the actual image (plus a little extra). So, it is possible for a Sony (and Nikon using Sony exmor sensors) to have a negative number, but lets get real, they probably define that as 00000000000000.

  8. #28
    Senior Member
    Join Date
    Dec 2008
    Location
    Planet Earth
    Posts
    3,110
    From the posts so far is it a conclusion that raising the shadows of the 5D IV 4-5 stops is not acceptable, but 3 might be.
    And
    Sony and Nikon are better for astrophotographers.
    And
    Daniel Browning no longer posts on this forum.

    We might go back and look at some of Daniel's complaints about Canon. But if my memory is correct this kind of summed up why we no longer see him posting.

  9. #29
    Super Moderator Kayaker72's Avatar
    Join Date
    Dec 2008
    Location
    New Hampshire, USA
    Posts
    5,565
    Ha! Very true about Daniel. Had some great, but very technical, posts (and also great pictures). But I learned a lot from him. His participation dropped off after he switched to Nikon because he was tired of the issues with Canon which weren't fixed with the 5DIII/1DX releases in 2012. But, one of the last posts of his that I remember he came back with a long list of issues he had with Nikon and their sensor tech (Sony exmor sensor).

    Grass is just different shades of green depending on where you are.

  10. #30
    Senior Member
    Join Date
    Dec 2008
    Location
    Central Kentucky
    Posts
    3,613
    Maybe different shades of green depending on your sensor

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •