86 replies to this topic

[alcyone00]

That's a really useful set of test files - thanks for making the effort!

 

Unfortunately it confirms beyond doubt that in-camera processing is producing rings in the raw data.  As before, I've debayered and binned the data (to reduce noise), then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. Here is the result after stretching:

 

CanonEosR8_Rings_RdivG.jpg

 

You can see a very clear progression of sharply defined concentric rings. Doing the same with the blue channel instead of the red channel gives the following:

 

CanonEosR8_Rings_BdivG.jpg

 

Something might be going on in the blue channel but I'm not sure what.

 

You've already said that you don't notice the rings and mottle at higher ISOs, so that's probably the best workaround.  Certainly with my EOS R there are no rings at ISO 1600 and above in a stacked image.

 

We now know for certain that both the EOS R/Ra and EOS R8 produce rings at low ISO.  There is therefore a possibility that all Canon mirrorless cameras are similarly affected.

 

Mark

Thank you for doing this, actually I see it on ISO 3200 too, maybe less but its visible, only time I didn't notice them was when I manually dither and ISO 3200, I tried adding flats but they didn't help much, there is a difference though. Yesterday I tried dithering and ISO 800 and they were visible again so I don't know anymore what has biggest impact on them.

 

Can this be considered as a sensor defect and is it possible that only some units are affected? I dont think warranty helps in this case because they will probably say nothing is wrong, I mean its visible only after a lot of stretching, I doubt they will see it as a reason for repair/replacement.

EOS R8 and R6 mk2 are using the same sensor but I didn't see anyone mentioning this problem, would be nice if we can get some r6 raw photos to test.

 

From your experience what is the best way to combat those rings if its even possible?

 

Here is ISO 3200 photo, if you boost saturation you can recognize ring pattern.

 

 

ISO 3200, 20 sec, no dithering. Link to RAW photos on google drive is few posts above.

 

 

[sharkmelley]

Here is ISO 3200 photo, if you boost saturation you can recognize ring pattern.

 

3200.jpg

 

ISO 3200, 20 sec, no dithering. Link to RAW photos on google drive is few posts above.

It certainly looks like the ring pattern.  I'll download the raws and report back.

[sharkmelley]

I downloaded 100 of your ISO 3200 lights and did a straight average.  As before, I debayered and binned the data (to reduce noise), then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. Here is the result after stretching:

 

 

I'm not totally convinced that we are seeing rings.  A better test would be to take 100 flats at ISO 3200 with the back of camera histogram in approximately the same position as the lights you have.

 

My experience with the EOS R suggests that the best way to combat the rings is to use an ISO of at least 800.  It may also be necessary to arrange for the back of camera histogram to be halfway across, which is a much higher exposure level than is generally recommended for astrophotography.  But this advice doesn't necessarily translate across to the R8.

 

Mark

[alcyone00]

I downloaded 100 of your ISO 3200 lights and did a straight average.  As before, I debayered and binned the data (to reduce noise), then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. Here is the result after stretching:

 

CanonEosR8_ISO3200_100lights_RdivG.jpg

 

I'm not totally convinced that we are seeing rings.  A better test would be to take 100 flats at ISO 3200 with the back of camera histogram in approximately the same position as the lights you have.

 

My experience with the EOS R suggests that the best way to combat the rings is to use an ISO of at least 800.  It may also be necessary to arrange for the back of camera histogram to be halfway across, which is a much higher exposure level than is generally recommended for astrophotography.  But this advice doesn't necessarily translate across to the R8.

 

Mark

Thank you, I will try to take those flat frames tomorrow. You also need lights from same session or flats with same ISO as lights from previous session is okay?

Edited by alcyone00, 12 September 2023 - 05:24 PM.

[sharkmelley]

Thank you, I will try to take those flat frames tomorrow. You also need lights from same session or flats with same ISO as lights from previous session is okay?

A set of 100 flats at ISO 3200.  I don't need lights but the flats should be exposed to roughly the same level as those ISO 3200 lights i.e. the peak of the back-of-camera histogram in approximately the same position.

[alcyone00]

A set of 100 flats at ISO 3200.  I don't need lights but the flats should be exposed to roughly the same level as those ISO 3200 lights i.e. the peak of the back-of-camera histogram in approximately the same position.

https://drive.google...?usp=drive_link

 

See if these are good, if not I will take another set. I used white shirt over lens pointing at white monitor screen...

[sharkmelley]

https://drive.google...?usp=drive_link

 

See if these are good, if not I will take another set. I used white shirt over lens pointing at white monitor screen...

Thanks for an excellent set of test data.  Unfortunately the test confirms that the red channel concentric rings do exist at ISO 3200.

 

I downloaded your100 ISO 3200 flats and did a straight average (floating point arithmetic).  As before, I debayered and binned the data (to reduce noise), then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. I also divided the (bias-subtracted) blue channel by the (bias-subtracted) green channel.  Here is the result after stretching:

 

 

There are definitely no rings in the blue channel (right-hand pane) but it looks like there could be rings in the red channel (left-hand pane) though they are camouflaged by the mottle.

 

I then took the earlier result from one of your ISO 100 flats (again red divided by green):

 

 

I chose this particular flat because the values in the red channel are broadly similar.

Since the background mottle is also very similar, it can be removed (i.e. calibrated out) by dividing the ISO 3200 result by the ISO 100 result and stretching:

 

 

All we are left with is the rings (no mottle) and it proves that the rings in the ISO3200 result have similar (but not identical) spacing to the rings in the ISO100 result.  They are not identical because the red values in the ISO 100 flat are slightly higher than the red values in the ISO 3200 flats and this affects the ring spacing.

 

The presence of rings across a broad range of ISO values is enormously frustrating for anyone wanting to use the Canon EOS R8 for deep-sky astrophotography.  I think the best way to reduce the effects of the rings is to use longer exposures at ISO 3200 to push the back-of-camera histogram further to the right, making the rings closer together and less conspicuous.  Unfortunately this will also saturate a great many stars and any bright nebulosity.

 

Mark

Edited by sharkmelley, 14 September 2023 - 09:02 AM.

[alcyone00]

Thanks for an excellent set of test data.  Unfortunately the test confirms that the red channel concentric rings do exist at ISO 3200.

 

I downloaded your100 ISO 3200 flats and did a straight average (floating point arithmetic).  As before, I debayered and binned the data (to reduce noise), then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. I also divided the (bias-subtracted) blue channel by the (bias-subtracted) green channel.  Here is the result after stretching:

 

CanonEosR8_ISO3200_100flats_RdivG.jpg

 

There are definitely no rings in the blue channel (right-hand pane) but it looks like there could be rings in the red channel (left-hand pane) though they are camouflaged by the mottle.

 

I then took the earlier result from one of your ISO 100 flats (again red divided by green):

 

CanonEosR8_ISO100flat_RdivG.jpg

 

I chose this particular flat because the values in the red channel are broadly similar.

Since the background mottle is also very similar, it can be removed (i.e. calibrated out) by dividing the ISO 3200 result by the ISO 100 result and stretching:

 

CanonEosR8_RdivG_ISO3200_div_ISO100.jpg

 

All we are left with is the rings (no mottle) and it proves that the rings in the ISO3200 result have similar (but not identical) spacing to the rings in the ISO100 result.  They are not identical because the red values in the ISO 100 flat are slightly higher than the red values in the ISO 3200 flats and this affects the ring spacing.

 

The presence of rings across a broad range of ISO values is enormously frustrating for anyone wanting to use the Canon EOS R8 for deep-sky astrophotography.  I think the best way to reduce the effects of the rings is to use longer exposures at ISO 3200 to push the back-of-camera histogram further to the right, making the rings closer together and less conspicuous.  Unfortunately this will also saturate a great many stars and any bright nebulosity.

 

Mark

Thanks for such deep analysis. Do you know any other model that has rings on such high ISO ? Also do you think a reason might be that EOS R8 has only electronic and electronic first curtain shutter, no mechanical. I used first curtain electronic shutter on all of my tests. If you want to do any other tests I can provide frames.

[sharkmelley]

Thanks for such deep analysis. Do you know any other model that has rings on such high ISO ? Also do you think a reason might be that EOS R8 has only electronic and electronic first curtain shutter, no mechanical. I used first curtain electronic shutter on all of my tests. If you want to do any other tests I can provide frames.

Only the EOS R/Ra, R5 and now the R8 have been properly tested as far as I know. The R5 has no rings at all and the R8 is the only one to show rings at both low ISO and high ISO.

 

There is a small possibility that the R8 has no rings using a fully electronic shutter but I doubt it.  I'm quite happy to analyse any exposures - I suggest you use the test protocol at ISO 100, mentioned earlier. 

Edited by sharkmelley, 14 September 2023 - 02:03 PM.

[alcyone00]

Only the EOS R/Ra, R5 and now the R8 have been properly tested as far as I know. The R5 has no rings at all and the R8 is the only one to show rings at both low ISO and high ISO.

 

There is a small possibility that the R8 has no rings using a fully electronic shutter but I doubt it.  I'm quite happy to analyse any exposures - I suggest you use the test protocol at ISO 100, mentioned earlier. 

Hi, here are 7x ISO 100 photos using electronic shutter and with different lens now. I also put ISO 6400 - 100 frames if you want to check and compare result with 3200 to see if there are improvements.

 

https://drive.google...ePh?usp=sharing

[sharkmelley]

Hi, here are 7x ISO 100 photos using electronic shutter and with different lens now. I also put ISO 6400 - 100 frames if you want to check and compare result with 3200 to see if there are improvements.

 

https://drive.google...ePh?usp=sharing

I downloaded your electronic shutter ISO100 flats.  As before, I debayered and binned the data (to reduce noise), then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. Here is the result after stretching:

 

 

You can see the progression in the rings as the level of exposure increases.

 

I also downloaded your 100 ISO6400 flats and did a straight average (floating point arithmetic).  Again, I debayered, binned then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. Here is the result after stretching:

 

 

I cannot detect any rings.  Maybe they are still present but, if so, they at such a low level that they are not detectable for practical purposes.  In any case this suggests a possible way forward for successful deep-sky imaging with the EOS R8 i.e. ISO6400 with the back of camera histogram at the halfway point.  This might also work with the electronic 1st curtain shutter.

 

Mark

[alcyone00]

I downloaded your electronic shutter ISO100 flats.  As before, I debayered and binned the data (to reduce noise), then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. Here is the result after stretching:

 

CanonEosR8_electronicshutter_ISO100flat_RdivG.jpg

 

You can see the progression in the rings as the level of exposure increases.

 

I also downloaded your 100 ISO6400 flats and did a straight average (floating point arithmetic).  Again, I debayered, binned then divided the (bias-subtracted) red channel by the (bias-subtracted) green channel. Here is the result after stretching:

 

CanonEosR8_electronicshutter_ISO6400_100lights_RdivG.jpg

 

I cannot detect any rings.  Maybe they are still present but, if so, they at such a low level that they are not detectable for practical purposes.  In any case this suggests a possible way forward for successful deep-sky imaging with the EOS R8 i.e. ISO6400 with the back of camera histogram at the halfway point.  This might also work with the electronic 1st curtain shutter.

 

Mark

Interesting, thank you. I will try some proper astro at iso 6400 when weather allows it.

[sharkmelley]

My experience with the EOS R suggests that the best way to combat the rings is to use an ISO of at least 800. 

It turns out that my advice of using a high ISO on the EOS R does not fix the problem.  I hit a serious ring problem at ISO 1600:

https://www.cloudyni...ents-and-rings/

[primeshooter]

Odd to see canon taking a backward step here. Generally speaking, they've been good for astro work across the board. Does this suggest it was never really intentional or have they just goofed with the R8 I wonder...

[Rlakjdlsj]

Post #34:  " Only the EOS R/Ra, R5 and now the R8 have been properly tested as far as I know. The R5 has no rings at all and the R8 is the only one to show rings at both low ISO and high ISO."

 

Is the Canon R8, the 'Lord of the Rings?  I don't think so.  Old english saying,  "the proof is in the pudding."   Where are the R8 rings in all these deep, dusty exposures with the R8???  Do we have to go on a quest to Mordor to find them?  ;  )   

 

https://app.astrobin...48/canon-eos-r8

Edited by Ron359, 15 December 2023 - 08:22 PM.

[alcyone00]

Post #14:  " Only the EOS R/Ra, R5 and now the R8 have been properly tested as far as I know. The R5 has no rings at all and the R8 is the only one to show rings at both low ISO and high ISO."

 

Is the Canon R8, the 'Lord of the Rings?  I don't think so.  Old english saying,  "the proof is in the pudding."   Where are the R8 rings in all these deep, dusty exposures with the R8???  Do we have to go on a quest to Mordor to find them?  ;  )   

 

https://app.astrobin...48/canon-eos-r8

Maybe i got defect sensor i dont know. There was software update few weeks ago and I didnt try astro after, will try in next few days when weather allows it but I kinda lost my will for astro after all this troubleshooting, I tried every possible combination of settings and rings would always appear. Only higher ISO and manual dithering every few frames would help so there are no rings/weird magenta/green patterns on the image. Only thing I didn't try was shooting from area with no light pollution but idk if that matters. Watching all those dusty images makes me wonder if all that dust can "hide" rings from the image. Also he used same lens as me (samy 135mm). I will try some images with canon 50mm this time and report back in a week or two.

[sharkmelley]

Post #14:  " Only the EOS R/Ra, R5 and now the R8 have been properly tested as far as I know. The R5 has no rings at all and the R8 is the only one to show rings at both low ISO and high ISO."

 

Hi Ron359,

 

I accept that I was incorrect when I made that comment (by the way it was post #34 not post #14).  Since then I have established that the R8 is not the only Canon full-frame mirrorless camera to produce rings at high ISO.  In fact my own EOS R can produce a ring (or rings) at ISO 1600 under certain circumstances.  See this thread for further information. 

 

Happily, my analysis of the underlying nature of the problem (in-camera digital scaling with a multiplicative value very close to unity) led to a solution, which can be seen in that thread.

 

By the way, I'm always happy to explain and defend my diagnostic techniques to anyone who shows a genuine interest.

[sharkmelley]

Maybe i got defect sensor i dont know. 

No, it's definitely not a defective sensor.  It's Canon monkeying around with your precious data.

[Rlakjdlsj]

Hi Ron359,

 

I accept that I was incorrect when I made that comment (by the way it was post #34 not post #14).  Since then I have established that the R8 is not the only Canon full-frame mirrorless camera to produce rings at high ISO.  In fact my own EOS R can produce a ring (or rings) at ISO 1600 under certain circumstances.  See this thread for further information. 

 

Happily, my analysis of the underlying nature of the problem (in-camera digital scaling with a multiplicative value very close to unity) led to a solution, which can be seen in that thread.

 

By the way, I'm always happy to explain and defend my diagnostic techniques to anyone who shows a genuine interest.

I corrected the typo..  Real images keep  made that don't support or reflect your test results.  

Edited by Ron359, 16 December 2023 - 12:47 AM.

[sharkmelley]

I corrected the typo.. Real images keep  made that don't support or reflect your test results.  

You risk denying the validity of both the experimental/mathematical work I have done and the continuing struggles of the original poster.  You also have another typo 

 

To the original poster and anyone else struggling with the Canon "rings" problem I have the following practical advice:

• Shoot both the lights and flats at ISO 1600 or higher. 
• Choose exposure lengths for the lights and flats to ensure the peak of the back-of-camera histogram is nearly halfway across.  But be aware that such exposures are much higher than is generally recommended and will saturate more stars.
• If possible, shoot flats with a light source whose brightness can be varied from exposure to exposure (the dusk sky is good for this).  This will dither the position of any potential rings in the frame since the ring positions vary according to brightness.

Hopefully you'll then be able to produce ring-free images of the quality that Ron359 pointed to earlier.

[Pappis]

Only the EOS R/Ra, R5 and now the R8 have been properly tested as far as I know. The R5 has no rings at all and the R8 is the only one to show rings at both low ISO and high ISO.

 

There is a small possibility that the R8 has no rings using a fully electronic shutter but I doubt it.  I'm quite happy to analyse any exposures - I suggest you use the test protocol at ISO 100, mentioned earlier. 

Do you know, if lens corrections were ON in the bodies tested? They are known to cause all kinds of artefacts when not needed.

 

Also, do you know which shutter mode was used in these tests?

[sharkmelley]

Do you know, if lens corrections were ON in the bodies tested? They are known to cause all kinds of artefacts when not needed.

Also, do you know which shutter mode was used in these tests?

In general the lens corrections are off but it doesn't seem to make any any difference to the digital scaling of the data.  Both electronic and manual shutter modes were tested - again it makes no difference.

[sharkmelley]

The above question from Pappis provoked me to take a another look at the cause of the red channel rings on the Canon EOS R8.  It quickly became clear that the effect was different to the EOS R/Ra.  For instance, the histogram of the R8 red channel does not display the spikes and notches of the R/Ra red channel, which would indicate channel scaling by a fixed multiplier.

 

A bit more analysis indicated very strongly that a radial quadratic function (i.e. quadratic in "r" where "r" is the distance from the image centre) is being applied to the red channel and the net effect is to increasingly brighten the red channel, the further it is from the image centre.  The magnitude of the effect is to brighten the periphery of the red channel by approx 2% in the case of the Canon EF 50mm f/1.8 lens and by approx 1% in the case of the Samyung 135mm lens (which is not recognised by the camera). In both cases "Peripheral Illumination Correction" was disabled.  My guess is that it's an undocumented colour shading correction.

 

If I'm right about this (obviously more testing would be required to absolutely confirm it) then it makes the artefact very similar in behaviour to the Nikon hardcoded correction.  In this linked webpage I describe how such quadratic corrections cause steps/rings in image data.

Edited by sharkmelley, 17 December 2023 - 04:53 AM.

[Pappis]

Thank you for your insight, sharkmelley! Assuming the cause for the need to apply radial color-dependent corrections is in the sensor OWB/UV/IR filter coatings, your findings can be understood pretty easily.

 

My assumption for Canon R8 was originally that the rings may be somehow connected to the amplification steps seen in the dynamic range and/or dynamic range shadow improvement curves measured by PhotonsToPhotos. I do not (yet) have the R8, but I was and still am planning to acquire one for astrophotography. If I had it already, I'd take flats (following your instructions) with ISO100/200/400/800/1600/3200/6400 to see, if the rings change with ISO.

 

But I assume, darks would likely correct the rings.

Edited by Pappis, 17 December 2023 - 05:08 AM.

[sharkmelley]

I think the purpose of the undocumented radial colour corrections is likely to be sensor crosstalk i.e. the problem where a photon entering the pixel of one colour ends up creating an electron in the storage well of an adjacent pixel of a different colour.  Here is a Samsung technical paper which mentions using polynomial corrections for crosstalk.

 

I think the Sony concentric colour polygons are also a correction for the same problem of crosstalk but instead of using a polynomial such as R-squared where R is the distance from the image centre, Sony uses an easier-to-calculate approximation which generates polygons instead of circles.  Bizarrely, Sony actually patented this polygon method!

 

Unfortunately your assumption that darks will correct the rings is incorrect.  Darks and flats are unable to correct them.