X-T2: Low Light ISO noise

[Kottan]

Hi,

I don't want to start a big discussion on noise here, just want to get a bit of feedback on this topic to correct my (high) expectations or get some hints to change my technique.

My general expectations on this camera have been met after some weeks of real use and I've settled with what I got - satisfied.

Still, after what I've read about high ISO performance of the Fujifilm X system I did expect to be able getting quite clean pictures (or maybe: a bit noisy but pleasingly grainy) in available light up to ISO 3200, maybe even ISO 6400.

What I realize in real use: I can't see any improvement over what I'm used to from any camera i owned - the higher the ISO the more details are destroyed, in a linear progression. Considering the facts concerning "isoless sensors" and discussions that there should be no difference in noise between ISO400 and ISO800 (up to 800) or ISO1000 versus ISO1600 (up to ISO1600) - I can't verify that on my X-T2: noise gets worse the higher ISO is set in a linear progression.

Also, I do see a lot of high ISO pictures (say 6400) on the web that seem to be quite clean at the 1st glance (cannot tell how those look when zooming in a bit). When I look at one of my pictures taken with ISO1600 in low light there just is more noise that I'd have expected with the X-T2. Noise gets worse the darker the scene gets. As a result, I find usable ISO is max. 1600 for my taste.

The contradiction over the opinions about ISO-noise is, that there are people who celebrate how good the performance at ISO12000, while others just consider everything over 1600 as unusable. Sure, judgements my be very subjective and dependent on the scene (simple vs. lot of details) on this topic, but opinions seem to differ a lot ...

So, can you enlighten me: did I expect too much? Am I doing something wrong?

Thanks,

Fred

 

[JacquesC]

High ISO noise is always a thorny issue with any camera, but here is what I've observed:

I cannot successfully process Fuji files for noise in the same way as my Nikon files, I have to do it differently, and I'm still struggling with that sometimes. You will get many opinions on how to process the Fuji files, and an equal number of opinions on which raw processor to use - so I'm not going into that.

For me I don't want to start using different pieces of software for different cameras, I'm using LR/PS and will stick to that for the time being. I very often get images shot at ISO's 6400 and even higher that are quite acceptable to me. Obviously if you look at it at 100% or higher you will see noise, but I take my intended purpose of an image into consideration before making a call on whether to accept or reject an image.

All in all I have to say that the high ISO noise from the X-T2 is not dramatically different from what I get with my D750 - about 1/2 - 1 stop difference.

What I would suggest though is that you shoot RAW+JPG and turn NR to 0 or even lower, and keep sharpening low. If you then look at the jpg output from the camera you might be surprised how clean those files look - very often I cannot match the cleanliness of the jpg output with my own processing of the raw files. The jpg OOC files from the X-T2 are heavily influenced by the sharpness and noise reduction settings, so you need to experiment with those.

Bottomline: You need to experiment and try various strategies to keep the noise levels acceptable. In my use and experience of the X-T2 high ISO noise has become a non-issue for me and hardly ever a factor to consider when determining which camera to use.

 

[Iuvenis]

I had a thread on a similar topic on the Science and Technology forum.

Some noise, shot noise, increases linearly based on the light hitting the sensor and independent of camera design (so for a given brightness on the image, based on ISO setting). Some noise is added to this in camera, but the amount of that noise has been decreasing with every sensor generation.

The result is that most noise nowadays is independent of camera design, though there still are variations. So called ISOless sensors are the same as any other sensor when it comes to shot noise, they just have more flexibility in post processing and often have superior dynamic range.

Because noise is linear, its visibility doubles when you double your output size. However, this is only apparent when noise is what limits your output size. The resolution of the sensor/lens combination limits print size as well.

The result is that at low iso settings, you will find you can print so large before noise becomes an issue, resolution becomes an issue first. Beyond a certain level (about iso 800 in my experience) noise starts to reduce the maximum size of print you can make, regardless of how good your technique and lens may be.

As a result, people do frequently say that iso is good 'up to' a certain level. They mean that for their proposed output size, it won't limit them. If your output size is very small, that may be up to a very high iso setting.

On the other hand, there's no point in worrying about the very low levels of noise at low iso settings. If you output large enough to see that noise, you'll encounter other problems first.

 

[lexvo]

Maybe you can post one or two examples so that we can see what you mean.

And don’t forget exposure is very important, especially when shooting high ISO. A slight underexposure can lead to a dispropotional increase in noise.

 

[Kottan]

What I would suggest though is that you shoot RAW+JPG and turn NR to 0 or even lower, and keep sharpening low. If you then look at the jpg output from the camera you might be surprised how clean those files look - very often I cannot match the cleanliness of the jpg output with my own processing of the raw files.

Thank you Jacques,

very interesting point - I do shoot RAW+JPG, but for the only reason that I can zoom in to max on the camera. Yes, the JPGs are very good, but I prefer RAW editing to get my pictures exactly as I want them. So JPG settings are not relevant, but I have set sharpening to 0 and NR to the lowest possible value.

I will definitely give the JPGs a try in low light, curious if the JPGs will be cleaner than the processed RAWs, where I apply sharpening an NR to my taste in post.

 

[Kottan]

Some noise, shot noise, increases linearly based on the light hitting the sensor and independent of camera design (so for a given brightness on the image, based on ISO setting). Some noise is added to this in camera, but the amount of that noise has been decreasing with every sensor generation.

Thanks Luvenis,

my findings are, that noise increases in principle linear with higher ISO, but not with light hitting the sensor - the more under exposure at a given ISO setting, the more noise ...

 

[DarnGoodPhotos]

What I would suggest though is that you shoot RAW+JPG and turn NR to 0 or even lower, and keep sharpening low. If you then look at the jpg output from the camera you might be surprised how clean those files look - very often I cannot match the cleanliness of the jpg output with my own processing of the raw files.

Thank you Jacques,

very interesting point - I do shoot RAW+JPG, but for the only reason that I can zoom in to max on the camera. Yes, the JPGs are very good, but I prefer RAW editing to get my pictures exactly as I want them. So JPG settings are not relevant, but I have set sharpening to 0 and NR to the lowest possible value.

I will definitely give the JPGs a try in low light, curious if the JPGs will be cleaner than the processed RAWs, where I apply sharpening an NR to my taste in post.

I also have a hard time matching the camera’s noise/sharpening results, but at ISO1600 it really shouldn’t be an issue. Remember that you clan easily under expose in camera by 2 stops and push the Raws two stops in post.

 

[Iuvenis]

Bear in mind that noise is most visible in the shadows. The shadows in your image have already been exposed to several stops less light than the mid-tones, even at base iso. That is why they are shadows.

If you underexpose a couple of stops and push in post processing, the noise will not be noticeable in the brighter parts of the image, but may well be visible in the shadows.

 

[Kottan]

Maybe you can post one or two examples so that we can see what you mean.

And don’t forget exposure is very important, especially when shooting high ISO. A slight underexposure can lead to a dispropotional increase in noise.

Thanks Lexvo,

I really would have liked to upload a pic to the forum/gallery here and there, but unfortunately this seems to be only possible when Flash is installed - Flash is banned from my computer, so there is no way it seems ...

You may find some low light shots of the X-T2 at ISO1000 on my flickr page, but these turned out quite good - the questionable picture won't find their way to Flickr. Here is the link: Kottan's Flickr

I did learn the lesson of exposure at high ISO at the very beginning of my Fujifilm experience. I've been used to underexpose frequently on my Nikons for various reasons, also in low light to reduce ISO. With the X-T2 the same trick didn't work, as noise got worse when underexposing as compared to right exposure at higher ISO. Finally I converted to auto exposure with the X-T2, as it does this job really well (as it does auto WB).

 

[John Motts]

the more under exposure at a given ISO setting, the more noise ...

This is to be expected with any camera.

When we talk about about ISO-less cameras what we mean is that the results from a shot taken at 200 ISO for example and underexposed by 5 stops will be pretty similar to a correctly exposed shot at 6400 ISO.

 

[stevo23]

Hi,

I don't want to start a big discussion on noise here, just want to get a bit of feedback on this topic to correct my (high) expectations or get some hints to change my technique.

My general expectations on this camera have been met after some weeks of real use and I've settled with what I got - satisfied.

Still, after what I've read about high ISO performance of the Fujifilm X system I did expect to be able getting quite clean pictures (or maybe: a bit noisy but pleasingly grainy) in available light up to ISO 3200, maybe even ISO 6400.

What I realize in real use: I can't see any improvement over what I'm used to from any camera i owned - the higher the ISO the more details are destroyed, in a linear progression. Considering the facts concerning "isoless sensors" and discussions that there should be no difference in noise between ISO400 and ISO800 (up to 800) or ISO1000 versus ISO1600 (up to ISO1600) - I can't verify that on my X-T2: noise gets worse the higher ISO is set in a linear progression.

I don't know if it really means "no difference in noise between ISO400 and ISO800". See my comment below.

Also, I do see a lot of high ISO pictures (say 6400) on the web that seem to be quite clean at the 1st glance (cannot tell how those look when zooming in a bit). When I look at one of my pictures taken with ISO1600 in low light there just is more noise that I'd have expected with the X-T2. Noise gets worse the darker the scene gets. As a result, I find usable ISO is max. 1600 for my taste.

ISO invariance seems to mean that a sensor will generally produce the same image quality when underexposing by a few stops at base ISO and brightening it in post as it would if you had captured at the "proper" ISO. I've never taken it to mean that ISO1600 properly exposed, will look just the same as ISO800 properly exposed.

On my XP2, I don't mind ISO1600 or ISO3200 if there are not a lot of smooth surfaces in an image. IE, the one below.

The contradiction over the opinions about ISO-noise is, that there are people who celebrate how good the performance at ISO12000, while others just consider everything over 1600 as unusable. Sure, judgements my be very subjective and dependent on the scene (simple vs. lot of details) on this topic, but opinions seem to differ a lot ...

I shot this at ISO3200 with my XP2 so you can compare to your own results:



ISO200 and ISO800 would have been cleaner. But I didn't have my tripod at the time.

So, can you enlighten me: did I expect too much? Am I doing something wrong?

Thanks,

Fred

I've not tested ISO-less with these sensors, I just know what it looks like at each setting and as I go up in ISO, the noise increases. I know there is a gain change at ISO800 and it's a very clean ISO800. But I don't have reasons to significantly underexpose images. So for me, ISO200 is always the standard if I can get away with it.

 

[Kottan]

When we talk about about ISO-less cameras what we mean is that the results from a shot taken at 200 ISO for example and underexposed by 5 stops will be pretty similar to a correctly exposed shot at 6400 ISO.

Yes John, I think I got that right - I just have made bad experience with underexposing with ISO set higher than 200, just to keep ISO lower than it should have been for a correct exposure. In my situation the results seemed to be better with higher ISO and correct exposure.

But I am at the beginning of this and will examine further for my needs ...

 

[Pritzl]

Some noise, shot noise, increases linearly based on the light hitting the sensor and independent of camera design (so for a given brightness on the image, based on ISO setting). Some noise is added to this in camera, but the amount of that noise has been decreasing with every sensor generation.

Thanks Luvenis,

my findings are, that noise increases in principle linear with higher ISO, but not with light hitting the sensor - the more under exposure at a given ISO setting, the more noise ...

Luvenis is correct. Shot noise is due to the random distribution of photons. WIth less light (whether through low light levels or underexposure) you get a lower signal-to-noise ratio. This results in the random speckles of darker and lighter spots we characterize as noise. Read noise is what sensor design might add to that and, for the most part, is no longer a major factor on modern sensors.

Next we come to ISO-less sensors, that does not mean that ISO has no effect on noise. It just means that the gain is so efficient it does not significantly worsen any existing shot noise. So, an ISO-less sensor will provide more or less the same results, at a given exposure, regardless of the ISO setting. It does not mean that higher ISO images will be just as clean as lower ISO ones globally.

The next frontier in sensor design will be increasing light sensitivity, e.g. better microlenses, full spectrum pixels, etc...

Finally, comparing a crop frame sensor's noise to that from a full frame camera is not fair. The latter benefits from a larger area which means 2 times the light is used to generate the image which naturally improves the signal to noise ratio (of the whole image, not on a per-pixel level).

--
http://www.flickr.com/photos/pritzl/

 

[Kottan]

I don't know if it really means "no difference in noise between ISO400 and ISO800". See my comment below.

Hi Stevo,

I don't know either, but remember reading, that ISO >200 to 800 and ISO >800 to 1600 should be quite consistent regarding IQ within either range. I for me cannot confirm that anyway ...

ISO invariance seems to mean that a sensor will generally produce the same image quality when underexposing by a few stops at base ISO and brightening it in post as it would if you had captured at the "proper" ISO. I've never taken it to mean that ISO1600 properly exposed, will look just the same as ISO800 properly exposed.

I'm not an expert by far, but I can see advantages when underexposing at base ISO vs higher ISO and correct exposure, but no more when underexposing ISO, say, >400. But I need to get into that more deeply ...

I shot this at ISO3200 with my XP2 so you can compare to your own results:

Thanks for the picture - I think it's at the level where I can get my pictures, too. Sure, it all depends from the scene and what is in the picture. It seems that my results are right there what you can get out of higher ISO - maybe I'm just too picky

I know there is a gain change at ISO800 and it's a very clean ISO800. But I don't have reasons to significantly underexpose images. So for me, ISO200 is always the standard if I can get away with it.

My standard too, if I have a tripod with me. Up to ISO800 it's pretty fine anyway. For my taste, ISO1600 is the limit - depends, of course. I've shot at ISO3200 yesterday in Vienna and wasn't very happy with the results at all - it was near to dark, and the noise in the sky was just too much. At the other day, I shot at ISO1000 in the night and the pictures turned out quite well. So ...

Let's take what we have and get the best out of it

Fred

 

[Kottan]

WIth less light (whether through low light levels or underexposure) you get a lower signal-to-noise ratio. This results in the random speckles of darker and lighter spots we characterize as noise.

Thanks Pritzl,

yes, that's the technical explanation for my observations

So, an ISO-less sensor will provide more or less the same results, at a given exposure, regardless of the ISO setting. It does not mean that higher ISO images will be just as clean as lower ISO ones globally.

Sorry, I don't understand: lower ISO images are cleaner, but at a given exposure ISO setting doesn't make a difference?

Finally, comparing a crop frame sensor's noise to that from a full frame camera is not fair.

Of course, this was not my point.

 

[woffles]

Maybe you can post one or two examples so that we can see what you mean.

And don’t forget exposure is very important, especially when shooting high ISO. A slight underexposure can lead to a dispropotional increase in noise.

Thanks Lexvo,

I really would have liked to upload a pic to the forum/gallery here and there, but unfortunately this seems to be only possible when Flash is installed - Flash is banned from my computer, so there is no way it seems ...

You may find some low light shots of the X-T2 at ISO1000 on my flickr page, but these turned out quite good - the questionable picture won't find their way to Flickr. Here is the link: Kottan's Flickr

I did learn the lesson of exposure at high ISO at the very beginning of my Fujifilm experience. I've been used to underexpose frequently on my Nikons for various reasons, also in low light to reduce ISO. With the X-T2 the same trick didn't work, as noise got worse when underexposing as compared to right exposure at higher ISO. Finally I converted to auto exposure with the X-T2, as it does this job really well (as it does auto WB).

In my experience, I've found that the XT2 tends to be conservative and underexpose to protect highlights so I have to push it some to get good exposures.

 

[Kottan]

In my experience, I've found that the XT2 tends to be conservative and underexpose to protect highlights so I have to push it some to get good exposures.

I switched from full manual exposure to auto exposure and for now I'm perfectly happy with matrix metering of the X-T2.

 

[Not you]

My experience with the XT2 when I compared the images to the XT1 (I shot with both at the same time for approx 1 year).

If I shot the same scene (in RAW) with both cameras at a higher ISO (i.e. 3200) I could see a bit more noise from the XT2. But ... when I decreased the size of the XT2 image to the same size as the XT1 image ... the XT2 image was better.

I never put much work or thought into ISO quality when images are down sampled for sharing online because the almost always look great when shrunk for web.

I believe one item that people are commenting on when they say the high ISO quality is great is they are referring to how much detail and colour is kept in the image compared to other sensors. The high ISO noise quality certainly differs from sensor to sensor and the Fuji may have just as much noise as a compared to a common Bayer sensor of the same generation ... but if the Fuji image has better detail and colour then that's a win for the Fuji.

I have shot weddings for 10 years and have worked with Nikon crop and full frame and now two Fuji XT2's. All camera have had their pro's and con's. But over the past few years shooting Fuji ... I can say when I'm printing a clients album and the low light shots are included (using no flash and a fast prime lens in a dim area such as the getting ready pics or during the ceremony or during the reception) ... those images print great. I have found that the way an image looks like on a 24" monitor is the way it would look printed at that size. But since most prints aren't that large, the high ISO images print great up to 8x12.

 

[stevo23]

I don't know if it really means "no difference in noise between ISO400 and ISO800". See my comment below.

Hi Stevo,

I don't know either, but remember reading, that ISO >200 to 800 and ISO >800 to 1600 should be quite consistent regarding IQ within either range. I for me cannot confirm that anyway ...

I don't think so personally.

ISO invariance seems to mean that a sensor will generally produce the same image quality when underexposing by a few stops at base ISO and brightening it in post as it would if you had captured at the "proper" ISO. I've never taken it to mean that ISO1600 properly exposed, will look just the same as ISO800 properly exposed.

I'm not an expert by far, but I can see advantages when underexposing at base ISO vs higher ISO and correct exposure, but no more when underexposing ISO, say, >400. But I need to get into that more deeply ...

I'm not sure I see the advantage.

I shot this at ISO3200 with my XP2 so you can compare to your own results:

Thanks for the picture - I think it's at the level where I can get my pictures, too. Sure, it all depends from the scene and what is in the picture. It seems that my results are right there what you can get out of higher ISO - maybe I'm just too picky

I know there is a gain change at ISO800 and it's a very clean ISO800. But I don't have reasons to significantly underexpose images. So for me, ISO200 is always the standard if I can get away with it.

My standard too, if I have a tripod with me. Up to ISO800 it's pretty fine anyway. For my taste, ISO1600 is the limit - depends, of course. I've shot at ISO3200 yesterday in Vienna and wasn't very happy with the results at all - it was near to dark, and the noise in the sky was just too much. At the other day, I shot at ISO1000 in the night and the pictures turned out quite well. So ...

Let's take what we have and get the best out of it

Fred

I've never used more than base ISO unless it's been absolutely necessary.

 

[Pritzl]

Sorry, I don't understand: lower ISO images are cleaner, but at a given exposure ISO setting doesn't make a difference?

If you take a picture at a set aperture, shutter speed and light conditions then the shot noise is the same, regardless of ISO setting. Changing ISO just changes the amplification (brightness) of the image. Older sensors introduced more noise (read noise) that can be worse at some ISO settings. Hence you got recommendations to avoid certain ISO settings for different cameras like skipping non-whole stops, etc...

With an ISO-less sensor, whether the picture is shot at ISO 200 and raised 4 stops in post or shot at ISO 3200, the noise is roughly the same. This allows you to underexpose (using a lower ISO) an image to preserve highlights, and then boost the shadows in post with minimal penalty. Used in this way, ISO-less sensors allow you to capture wider DR images.

Here are 2 DPR articles that you might find useful:

https://www.dpreview.com/articles/8...e-shedding-some-light-on-the-sources-of-noise

https://www.dpreview.com/articles/7450523388/sony-alpha-7r-ii-real-world-iso-invariance-study

tldr; expose to the right, (widest aperture and longest shutter speed you can get away with) preserving highlights, and then tone-map as required in post. ISO is less relevant with a modern sensor.

--
http://www.flickr.com/photos/pritzl/