For the A7r iii and iv, what is the trade off between shooting underexposed vs higher ISO with proper exposure? I know noise becomes an issue with higher ISOs in low light, but what is the realistic break even point when using exposure compensation and then bringing up the exposure in post to avoid the noise introduced by under exposing too much? The sensors have good DR, but there will still be a point where noise from underexposing will exceed that of proper exposure with higher ISO.
Last year I was taking photos at the circus with my a7r iii and wanted to keep the shutter speed around 1/250 to try to combat the blur from moving subjects. I was using -1 EV of exposure compensation and my ISO was around 640. I keep reading that The A7r iv will need a faster shutter speed to achieve the same results in sharpness. If I use -2 EV my shutter speed will get me to 1/500 but I wonder how much I can pull out of the image to get a proper exposure without negative effects.
This site I've found is the best resource for the measurements to help you determine it. Here's the article explaining the chart you should be looking at, and I believe the A7R IV was recently added, but I'm on mobile at the moment
saxguy wrote:
I keep reading that The A7r iv will need a faster shutter speed to achieve the same results in sharpness.
This has been blown out of proportion. If you want the same overall image sharpness, you can use exactly the same shutter speed. Even if you want the same 100% crop sharpness (vs A7R3), you only need about a 20% faster shutter, which is about 0.26 stops.
I'm a bit confused as to what those charts are saying. Is that saying that ISO is better below 320, and then after that just use post processing for better results?
buffalowolff wrote:
I'm a bit confused as to what those charts are saying. Is that saying that ISO is better below 320, and then after that just use post processing for better results?
Any step up means there is an improvement in noise and dynamic range, from switching to a circuit specifically for higher gain. You should aim for the most improvement as long as you don't clip highlights.
Any flat region means there's no improvement in noise or dynamic range, what some refer to as ISO invariant or ISOless. So you can under expose by setting your ISO to the leftmost part of the flat region you are in without penalty.
Bill's graph is an excellent resource but keep in mind it's a quantitative-only measurement. It doesn't account for the perceptual issues sometimes seen when pushing low-ISO images, including color casts from slight noise imbalances between color channels. Dpreview's ISO invariance widget is good for that. Here's a link for the A7rIII: https://tinyurl.com/yylh4afy
snapsy wrote:
Bill's graph is an excellent resource but keep in mind it's a quantitative-only measurement. It doesn't account for the perceptual issues sometimes seen when pushing low-ISO images, including color casts from slight noise imbalances between color channels. Dpreview's ISO invariance widget is good for that. Here's a link for the A7rIII: https://tinyurl.com/yylh4afy
snapsy wrote:
Bill's graph is an excellent resource but keep in mind it's a quantitative-only measurement. It doesn't account for the perceptual issues sometimes seen when pushing low-ISO images, including color casts from slight noise imbalances between color channels. Dpreview's ISO invariance widget is good for that. Here's a link for the A7rIII: https://tinyurl.com/yylh4afy
While it exists, it's extremely small and can be corrected with a tiny bump in color noise reduction in Lightroom.
Just remember to compare within one ISO invariant region, so 800-6400 ISO
Eruditass wrote:
While it exists, it's extremely small and can be corrected with a tiny bump in color noise reduction in Lightroom.
Just remember to compare within one ISO invariant region, so 800-6400 ISO
Shadow color casts can't be removed with color noise reduction. They can only be ameliorated with the shadow tint color slider, which works well in some photos but not others because it adjusts the entire color balance of the shadows.
snapsy wrote:
Shadow color casts can't be removed with color noise reduction. They can only be ameliorated with the shadow tint color slider, which works well in some photos but not others because it adjusts the entire color balance of the shadows.
Seems I misunderstood what exactly you're referring to, can you provide some screenshots of the areas you're referring to?
I am a little confused by Bill's graph, especially the range of the A9. It looks as if it's dramatically better above 640 than both the r3 and r4. Is that correct?
Note that shadow color casts on recent Sony bodies are very minor.
You need to compare within one ISO invariant region, so 800+3 up to 6400 ISO. Anything below 800 ISO is know to have worse performance as explained by the graph linked. With that, the color casts is minimal and all there exists is slight color noise blotchiness, from what I see.
Michael Everet wrote:
I am a little confused by Bill's graph, especially the range of the A9. It looks as if it's dramatically better above 640 than both the r3 and r4. Is that correct?
This is an improvement graph, so comparing to a cameras base ISO dynamic range. One of A9's criticism is it's low ISO dynamic range performance, so it's starting point and point of comparison is worse. By high ISO it's dynamic range is much more in line with everything else.
Eruditass wrote:
This is an improvement graph, so comparing to a cameras base ISO dynamic range. One of A9's criticism is it's low ISO dynamic range performance, so it's starting point and point of comparison is worse. By high ISO it's dynamic range is much more in line with everything else.
Eruditass wrote:
You need to compare within one ISO invariant region, so 800+3 up to 6400 ISO. Anything below 800 ISO is know to have worse performance as explained by the graph linked. With that, the color casts is minimal and all there exists is slight color noise blotchiness, from what I see.
If you see some at 800+3 I'd like to see.
Color casts are not from additional noise but from noise that is unbalanced between color channels, thus it is not reflected in quantitative noise measurements. They usually go hand-in-hand however since the imbalances occur at the deepest shadows just above clipping, which are reflected in base ISO or near base ISO pushes. As I indicated the color casts on recent Sony bodies is very minor. The blotchiness you see is from the granular tonality available at the lowest DN levels after they're pushed.
snapsy wrote:
Color casts are not from additional noise but from noise that is unbalanced between color channels, thus it is not reflected in quantitative noise measurements. They usually go hand-in-hand however since the imbalances occur at the deepest shadows just above clipping, which are reflected in base ISO or near base ISO pushes. As I indicated the color casts on recent Sony bodies is very minor. The blotchiness you see is from the granular tonality available at the lowest DN levels after they're pushed.
You could easily quantify such color casts, they just aren't explicitly quantified in this particular monochromatic metric. Nevertheless, as I implied they are very correlated, and following this graph is typically a very good indication of such noise, unless you have any examples of color casts within an ISO invariant region.
Also I don't understand what you mean by it not being easily correctable. The casts are typically the same color and brightness range, so you can easily target then with RGB curves.
Eruditass wrote:
You could easily quantify such color casts, they just aren't explicitly quantified in this particular monochromatic metric. Nevertheless, as I implied they are very correlated, and following this graph is typically a very good indication of such noise, unless you have any examples of color casts within an ISO invariant region.
You're right, you can quantify color casts but not as easily because they occur as fractional or near-factional DN values at the shadow clipping. And I was speaking specifically to the quantitative measurement in that ISO-invariant graph. And they're not correlated at all to aggregate invariant noise measurements, ie you can could have two sensors with identical ISO variant aggregate noise levels but one with color casts and another without.
Also I don't understand what you mean by it not being easily correctable. The casts are typically the same color and brightness range, so you can easily target then with RGB curves.
You've indicated that before and it's still incorrect. The casts can't be corrected with simple RGB curves because they're fractional values and non-linear with respect to their post-gamma tonal output. This is why the ACR/LR shadow tint slider exists. Due to that non-linearity attempting to fix the shadow casts with RGB curves will result in the creation of casts further up the tonal scale. Even the targeted shadow-tint slider often introduces unwanted chroma shifts elsewhere in the tonal range.
Here's an example of color casts on the Z6, which are introduced when using its 12-bit raw mode (which introduces fractional imprecision of DN values near shadow clipping):
snapsy wrote:
You're right, you can quantify color casts but not as easily because they occur as fractional or near-factional DN values at the shadow clipping. And I was speaking specifically to the quantitative measurement in that ISO-invariant graph. And they're not correlated at all to aggregate invariant noise measurements, ie you can could have two sensors with identical ISO variant aggregate noise levels but one with color casts and another without.
My point about the graph's metric correlating with the casts is referring to one sensor: when you will be likely to get casts, not comparing two sensors performance.
snapsy wrote:
You've indicated that before and it's still incorrect. The casts can't be corrected with simple RGB curves because they're fractional values and non-linear with respect to their post-gamma tonal output. This is why the ACR/LR shadow tint slider exists. Due to that non-linearity attempting to fix the shadow casts with RGB curves will result in the creation of casts further up the tonal scale. Even the targeted shadow-tint slider often introduces unwanted chroma shifts elsewhere in the tonal range.
Here's an example of color casts on the Z6, which are introduced when using its 12-bit raw mode (which introduces fractional imprecision of DN values near shadow clipping):
The issue is that such a shadow tint slider is a very coarse tool, even if it's earlier in the pipeline. A 3D LUT on the pre-demosaic'd RAW values would be best, but an RGB curve seems better than a simple slider honestly. Playing around with that and the color noise smoothness at least makes the shadows look non-tinted, but obviously I'd want to verify with a full color checker, and at that rate might as well do a 3D LUT.
