A7 III noise hot pixels on badly underexposed images in the shadows.

[IvankoPetro]

I have made a small test on A7 III to compare the ISO-less functionality

• Very underexposed image 1/800s ISO 100 F/16 then boosted exposure +5.5 EV
• And properly exposed image 1/1250s ISO 12,800 F/16

Both images virtually same in the noise department except one thing i don't quite understand in the shadows of ISO 100 image there's color dots like hot pixels? of different colors mostly around the vegetation in the lower left part of image.

At first i thought its because less light, but realistically both shots were done on F/16 so sensor seen same amount of light, there must be some technical explanation why there's more hot pixels? on ISO 100 (under exposed) shot.

There might be some hot pixel algorithm in the camera that does not fire up properly when scene is so under exposed or its not hot pixels but moire?

Source files: https://www.upload.ee/files/18669070/ISO_Test.zip.html

P.S. Used electronic shutter in both images if that matters.

 

[A74Me]

and ?

 

[TimoKoo]

I have made a small test on A7 III to compare the ISO-less functionality

There's a dual gain sensor in Sony 7AIII. It is Isoless only in ISO scales from 100 to 500 and from 640 to 51200. If you compare ISO 100 shot to ISO 12800 shot you are not doing the comparison in cameras ISOless range.
Look at this: https://www.photonstophotos.net/Charts/PDR.htm#Sony%20ILCE-7M3

• Very underexposed image 1/800s ISO 100 F/16 then boosted exposure +5.5 EV
• And properly exposed image 1/1250s ISO 12,800 F/16

Why are these pictures black and white? How did you boost the exposure?
I see red-magenta artefacts in both pictures, more visible in ISO 100 shot.

Both images virtually same in the noise department except one thing i don't quite understand in the shadows of ISO 100 image there's color dots like hot pixels? of different colors mostly around the vegetation in the lower left part of image.

At first i thought its because less light, but realistically both shots were done on F/16 so sensor seen same amount of light, there must be some technical explanation why there's more hot pixels? on ISO 100 (under exposed) shot.

The same aperture but different exposure times. The ISO 12800 shot got less light. (1/800 sec vs. 1/1250 sec.)

I think you should compare pictures with same exposure (same aperture and same exposure time).

There might be some hot pixel algorithm in the camera that does not fire up properly when scene is so under exposed or its not hot pixels but moire?

Source files: https://www.upload.ee/files/18669070/ISO_Test.zip.html

P.S. Used electronic shutter in both images if that matters.

Edit. I somehow lost the first picture in this post. Look at OP.

 

[IvankoPetro]

Since exposure time almost similar realized that i made ISO 100 photo probably at F/1.4 and ISO 12,800 photo at F/16 so that my mistake in original observations.

I have made a small test on A7 III to compare the ISO-less functionality

There's a dual gain sensor in Sony 7AIII. It is Isoless only in ISO scales from 100 to 500 and from 640 to 51200. If you compare ISO 100 shot to ISO 12800 shot you are not doing the comparison in cameras ISOless range.
Look at this: https://www.photonstophotos.net/Charts/PDR.htm#Sony%20ILCE-7M3

What do you mean ISO 12,800 is two EV below ISO 51,200 so well within range for iso-less functionality of sensor.

The same aperture but different exposure times. The ISO 12800 shot got less light. (1/800 sec vs. 1/1250 sec.)

I think you should compare pictures with same exposure (same aperture and same exposure time).

I think exposure difference is not as big to cause any effects its about 0.5 EV difference considering i am boosting 5.5 stops in the post processing.

A74Me, post: 68466923, member: 2056541"]
and ?

Noise looks quite annoying around the trees looks like Christmas tree lights, want to understand why its cleared when i select higher ISO.



[/QUOTE]

 

[TimoKoo]

Since exposure time almost similar realized that i made ISO 100 photo probably at F/1.4 and ISO 12,800 photo at F/16 so that my mistake in original observations.

A big difference to those numbers you gave in your first post. What is true?

In you OP you said:

• Very underexposed image 1/800s ISO 100 F/16 then boosted exposure +5.5 EV
• And properly exposed image 1/1250s ISO 12,800 F/16

I have made a small test on A7 III to compare the ISO-less functionality

There's a dual gain sensor in Sony 7AIII. It is Isoless only in ISO scales from 100 to 500 and from 640 to 51200. If you compare ISO 100 shot to ISO 12800 shot you are not doing the comparison in cameras ISOless range.
Look at this: https://www.photonstophotos.net/Charts/PDR.htm#Sony%20ILCE-7M3

What do you mean ISO 12,800 is two EV below ISO 51,200 so well within range for iso-less functionality of sensor.

ISO 12,800 is inside the Isoless range from ISO640 to ISO 51,200. ISO100 is out of that range. So, comparing ISO100 shot to ISO12800 shot you are not comparing Isoless things.

Why are these pictures black and white? How did you boost the exposure?
see red-magenta artefacts in both pictures, more visible in ISO 100 shot.

Reason i made it B/W so its more visible, my idea was if someone wants to check they download raw from link in first post.

A74Me, post: 68467038, member: 2227948"]
and ?

Noise looks quite annoying around the trees looks like Christmas tree lights, want to understand why its cleared when i select higher ISO.

It's the dual gain thing. Look at the Photons to Photos link.

But those red-magenta artefacts are obvious in your high Iso shot too.

 

[Bernard Delley]

Hot pixels do not go away usually. But, at high ISO they get clipped at the full well for that ISO. The high ISO clipping is not that much higher than tha signal as it is in underexposed low ISO.

You might look at the same scene at 100% pixel view, best wit a trusted raw conversion, for a series of of takes from very low ISO lifted in post to adequate ISO with the same exposure time and aperture.

 

[John Sheehy]

I have made a small test on A7 III to compare the ISO-less functionality

• Very underexposed image 1/800s ISO 100 F/16 then boosted exposure +5.5 EV
• And properly exposed image 1/1250s ISO 12,800 F/16

Both images virtually same in the noise department except one thing i don't quite understand in the shadows of ISO 100 image there's color dots like hot pixels? of different colors mostly around the vegetation in the lower left part of image.

At first i thought its because less light, but realistically both shots were done on F/16 so sensor seen same amount of light, there must be some technical explanation why there's more hot pixels? on ISO 100 (under exposed) shot.

There might be some hot pixel algorithm in the camera that does not fire up properly when scene is so under exposed or its not hot pixels but moire?

Source files: https://www.upload.ee/files/18669070/ISO_Test.zip.html

P.S. Used electronic shutter in both images if that matters.

What you implied about a threshold for outlier-pixel or mapped-bad-pixel rejection is certainly a possibility, but on the other hand, there is also the possibility that it is due to quantization noise in the raw file or in the converter, because "ISO-less-ness" or "ISO invariance" is usually a myth, literally, and is only practically true for common purposes, like mild "under-exposure", especially with high-key scenes. The only cameras where there is literally no difference in absolute noise at all at a range of ISOs, is when that range of ISOs all come from the same analog gain and digitization.

Our eyes average out fine detail and noise to some degree, but the pixels on our monitors are generally so big that we do see individual pixels to some extent, when they are not part of a tight dither pattern or repeating pattern which drives individual pixels into extinction. So, while we can expect over-quantization to mostly be cancelled out because it drives some pixels higher in value, while driving an equal number lower, we can still see the stronger-outlier effect of overly-quantized pixels.

Also, even before digitization and any resulting quantization,if the analog gain is different between two ISOs, then the absolute intensity of post-gain (late stage) read noise can not possibly be exactly the same.

 

[John Sheehy]

Since exposure time almost similar realized that i made ISO 100 photo probably at F/1.4 and ISO 12,800 photo at F/16 so that my mistake in original observations.

Never use different apertures when you don't have to. Sensors, especially BSI ones with small-ish pixels, lose light at a higher rate at very low f-numbers. So, you have a case where a camera is likely to multiply the original raw data by some value like 1.1 to 1.6 in the raw file.

If you're trying to isolate one characteristic, never vary multiple parameters. Exposure time should be the go-to variable for all comparison tests, as the only time it makes any difference is when exposure is long and dark current noise swells. In the case of your test, exposure should not have been varied at all.

Noise looks quite annoying around the trees looks like Christmas tree lights, want to understand why its cleared when i select higher ISO.

Analog gain gains over low-signal artifacts. It's like being 6 feet tall walking through leaf litter compared to being six inches tall and walking through the same leaf litter. When people talk about ISO-invariance, they are generally talking about the difference between being 6 feet tall or 60 feet tall; not about the range where lack of gain can give large differences in results.

 

[IvankoPetro]

Since exposure time almost similar realized that i made ISO 100 photo probably at F/1.4 and ISO 12,800 photo at F/16 so that my mistake in original observations.

Never use different apertures when you don't have to. Sensors, especially BSI ones with small-ish pixels, lose light at a higher rate at very low f-numbers. So, you have a case where a camera is likely to multiply the original raw data by some value like 1.1 to 1.6 in the raw file.

If you're trying to isolate one characteristic, never vary multiple parameters. Exposure time should be the go-to variable for all comparison tests, as the only time it makes any difference is when exposure is long and dark current noise swells. In the case of your test, exposure should not have been varied at all.

By exposure you mean the aperture?
I realized it was my mistake was too tired when tested now i retested in with F/16 both images i not getting anything close to 5EV range when shoot at ISO 100 and boost shadows in post processing. Got to do another test, but i am too tired at moment.

Anyway here i found a solution for hot pixels in RawTherapee this works like charm.

 

[Richard Butler]

What you implied about a threshold for outlier-pixel or mapped-bad-pixel rejection is certainly a possibility, but on the other hand, there is also the possibility that it is due to quantization noise in the raw file or in the converter, because "ISO-less-ness" or "ISO invariance" is usually a myth, literally, and is only practically true for common purposes, like mild "under-exposure", especially with high-key scenes.

The reason we settled on the term ISO invariance was because ISO-less or ISO-less-ness sounds like an absolute. We didn't want to be trying to write about something be "fairly ISO-less."

ISO invariance seemed to give some leeway to discuss the degree to which things exhibit variant or invariant behaviour. And I agree that there's no such thing as an absolutely ISO invariant sensor, even if some dual conversion gain sensors get pretty close, once at the raised gain step.

Richard - DPReview.com

 

[TimoKoo]

Since exposure time almost similar realized that i made ISO 100 photo probably at F/1.4 and ISO 12,800 photo at F/16 so that my mistake in original observations.

Never use different apertures when you don't have to. Sensors, especially BSI ones with small-ish pixels, lose light at a higher rate at very low f-numbers. So, you have a case where a camera is likely to multiply the original raw data by some value like 1.1 to 1.6 in the raw file.

If you're trying to isolate one characteristic, never vary multiple parameters. Exposure time should be the go-to variable for all comparison tests, as the only time it makes any difference is when exposure is long and dark current noise swells. In the case of your test, exposure should not have been varied at all.

By exposure you mean the aperture?

I can't answer what John Sheehy thinks about exposure. But he did not answer. I think the common way to think the exposure is to think it as a combination of the aperture, exposure time and scene lighting.

As John said: "exposure should not have been varied at all". If you want to test things, you should keep all three parts of exposure fixed. Then slowly, gradually change only one of those variables and then compare to the starting point.

ISO is not a part of exposure. But it's interesting what different Iso setting do. In OP you were interested in what effects different ISOs make to your picture. To test that you should keep the exposure (aperture, exposure time, lighting) fixed. Then take a series of pictures with diffrent ISOs and compare those together.

I realized it was my mistake was too tired when tested now i retested in with F/16 both images i not getting anything close to 5EV range when shoot at ISO 100 and boost shadows in post processing. Got to do another test, but i am too tired at moment.

Anyway here i found a solution for hot pixels in RawTherapee this works like charm.

I usually also check the Dead pixel box in RT.

 

[TimoKoo]

Since exposure time almost similar realized that i made ISO 100 photo probably at F/1.4 and ISO 12,800 photo at F/16 so that my mistake in original observations.

Never use different apertures when you don't have to. Sensors, especially BSI ones with small-ish pixels, lose light at a higher rate at very low f-numbers. So, you have a case where a camera is likely to multiply the original raw data by some value like 1.1 to 1.6 in the raw file.

If you're trying to isolate one characteristic, never vary multiple parameters. Exposure time should be the go-to variable for all comparison tests, as the only time it makes any difference is when exposure is long and dark current noise swells. In the case of your test, exposure should not have been varied at all.

By exposure you mean the aperture?

I can't answer what John Sheehy thinks about exposure. But he did not answer. I think the common way to think the exposure is to think it as a combination of the aperture, exposure time and scene lighting.

As John said: "exposure should not have been varied at all". If you want to test things, you should keep all three parts of exposure fixed. Then slowly, gradually change only one of those variables and then compare to the starting point.

ISO is not a part of exposure. But it's interesting what different Iso setting do. In OP you were interested in what effects different ISOs make to your picture. To test that you should keep the exposure (aperture, exposure time, lighting) fixed. Then take a series of pictures with diffrent ISOs and compare those together.

I realized it was my mistake was too tired when tested now i retested in with F/16 both images i not getting anything close to 5EV range when shoot at ISO 100 and boost shadows in post processing. Got to do another test, but i am too tired at moment.

Anyway here i found a solution for hot pixels in RawTherapee this works like charm.

I usually also check the Dead pixel box in RT.

I did download your pictures and played a little with them in RawTherapee. Is RT the Raw conversation tool of your choice?

Or do you use another raw converter usually, if so what? I ask because I think it is more important to learn to use your tools than jump from program to program and never learn to use one well.

It looks like you did not use any of denoise tools. There's many of them in RT. In the RT's Detail tab I suggest the Noise reduction tool and in this case also the Impulse Noise reduction in the Detail tab.

If you do not use RT, it's better to learn how to do the important things in your favorite raw converter.


A crop of ISO100 picture.


A crop of ISO12800 picture.

No big difference between these two pictures. I left a bit of noise in those to avoid plasticky look.

 

[IvankoPetro]

I did download your pictures and played a little with them in RawTherapee. Is RT the Raw conversation tool of your choice?

Or do you use another raw converter usually, if so what? I ask because I think it is more important to learn to use your tools than jump from program to program and never learn to use one well.

It looks like you did not use any of denoise tools. There's many of them in RT. In the RT's Detail tab I suggest the Noise reduction tool and in this case also the Impulse Noise reduction in the Detail tab.

If you do not use RT, it's better to learn how to do the important things in your favorite raw converter.

I mainly use Adobe Camera Raw (ACR), but it has no hot pixel correction algorithm offered by RawTherapee, i don't mind using both on same file for different tasks.

e.g. Photoshop ACR for Denoise AI then save TIFF and use RawTherapee for removing bad pixels, then open in Photoshop ACR again for final edit.

Anyway this test i did in original post was flawed i confused camera settings so besides hot pixels there's no conclusions to be made from the RAW's uploaded in first message.

 

[TimoKoo]

I did download your pictures and played a little with them in RawTherapee. Is RT the Raw conversation tool of your choice?

Or do you use another raw converter usually, if so what? I ask because I think it is more important to learn to use your tools than jump from program to program and never learn to use one well.

It looks like you did not use any of denoise tools. There's many of them in RT. In the RT's Detail tab I suggest the Noise reduction tool and in this case also the Impulse Noise reduction in the Detail tab.

If you do not use RT, it's better to learn how to do the important things in your favorite raw converter.

I mainly use Adobe Camera Raw (ACR), but it has no hot pixel correction algorithm offered by RawTherapee, i don't mind using both on same file for different tasks.

There is a hot pixel filter in ACR/LR but you don't see it because it is always on and you can't tweak it's effect by yourself.

e.g. Photoshop ACR for Denoise AI then save TIFF and use RawTherapee for removing bad pixels, then open in Photoshop ACR again for final edit.

Nice idea but it doesn't work. ;-) You can adjust a tiff picture in RT many ways but not use hot/dead pixel filter. Like other tools in RT's Raw Tab hot pixel correction only works with Raw files.

My suggest is to start ARW file's editing in ACR doing all your edits you like, with or without AI, and then it in PhotoShop. If there's still hot or dead pixels left you can use PS filter "Dust and Scratches" to get rid off hot pixels.

That is a good filter to clean dots but not so good to clean scratches. Start with small, maybe one px radius and high 50 to 100 threshold. Take care of details. It is easy to loose them.

Anyway this test i did in original post was flawed i confused camera settings so besides hot pixels there's no conclusions to be made from the RAW's uploaded in first message.

Like I said.

 

[jiberlin]

These are two different issues. Besides this, the images posted are much to small to see anything.