Leica M sensor low light performance state of the art?

[graham_mitchell]

Advertisement (gone after registration)

The problem with these type of figures -including DXO- is that they are not taken from the pure sensor output but from the raw data which include pre-raw noise reduction by the firmware, which will vary per camera.

 

Yes, good point.

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[albireo_double]

Can you even make an edge mask in Lightroom? I have not seen that tool.

 

Just hold the Alt key while sliding the Masking slider...or am I missing something?

[douglasf13]

In the article several different aspects of noise reduciton are discussed, so what we are taking here about is noise reduction involving different pixels, "in-camera filtering of raw data". The article by Emil Martinec shows and the post #23 by MikeE show that some cameras apply this type of noise reduction (Sony A700) other do not (Leica M 240 and Canon 5D II), you can clarly see it in the 2D- Fourier transform. It think it is quite a strong statement saying that all, even the most recent cameras with CMOS apply this sort of noise reduction. The two examples above show that this is not the case.

 

Thomas

 

I thought it should be mentioned that, after the A700's release (it was the first of the EXMOR sensor cameras,) there was such a backlash to the NR on the raws that Sony eventually released a firmware to allow the NR to be turned off.

 

Like Jim mentioned about the Nex-7, there seems to occasionally be a camera that will add a little "special sauce" at very high ISOs, but, now days, it is pretty rare that there is any in-camera, software NR baked into camera raws across the ISO range, outside of long exposure NR. Sony has been able to throw so much money and development into EXMOR over the last several years, since they sell the sensors to so many different companies, that they're currently at the cutting edge of commercial camera CMOS design.

 

Either way, the M240 seems a nice SNR and DR improvement over the M9 and not all that far off from EXMOR, which is what most asked for. It's a little optimistic to expect a smaller sensor company to compete with a behemoth like Sony, for better or worse.

[JimKasson]

I created a model for the noise floor behavior of the M240. This kind of model has sucessfully predicted the behavior of many cameras, but it can't explain the strangely low noise floor in the M240 at ISOs 200, 400, and 800.

 

I thought that pre- and post-gain offsets might explain what is going on, but I can't find any combination that works well. It's like there are two different modes of operation in the M240, one from 200 to 800, and one from 1600 to 3200. Is this the result of a deliberate design decision?

 

Characterizing the Leica M240, part 19 | The Last Word

 

Suggestions are welcome.

 

Thanks,

 

Jim

[jaapv]

The practical implications of this noise floor anomaly you found, Jim? Does it mean the underexpose-push technique is particularly effective up to ISO 800?

[JimKasson]

The practical implications of this noise floor anomaly you found, Jim? Does it mean the underexpose-push technique is particularly effective up to ISO 800?

 

I'm afraid that it might possibly be at the root of the push-processing green shift. It's hard to tell until I understand it better. It is strange behavior, especially for a camera from a manufacturer that is known for not doing much processing of raw images.

 

Jim

[JimKasson]

Advertisement (gone after registration)

The practical implications of this noise floor anomaly you found, Jim? Does it mean the underexpose-push technique is particularly effective up to ISO 800?

 

It's early days, but for now, for me, it means the opposite. I won't push ISO 800 and below images until I know what's going on.

 

Jim

[huckles]

Hi, thought I would put my own experience with my testing, although no where near as techincal. Just observation of the M at different ISO's pushed or native vs Nikon D4 push and native.

 

Up until 800, the Nikon D4 is 'just' slightly cleaner, but the M is doing very well. At 1600+ on the M this is where the D4 streaks ahead. You can push an M 800iso file 2 stops but that's when banding occurs. It certainly does have a huge colour shift to green though. ISO 3200 native on the M is much more accurate colours (and brighter!) but exhibits the same amount of banding. Not terrible banding, but not ideal. As a comparison, you can have any ISO range in the D4 and push it 4 stops before ANY banding is visible.

 

In the real world shooting with the M, absolutely the highest ISO I would use, unless it is an emergency is ISO 1600. Ideally ISO 800 with a slight push and correction of the green cast.

 

This is a very dissapointing find as if used this way i.e. 800 iso and then push. There is not much technological advantage over the M9 640 push is there... :S

 

After thinking about why Leica would change to M, it has become clear to me.. in the bid to add 'video', they have gone with CMOS.

 

I love my M, I loved the M9P... I'm the type of guy that just likes to know the real world benefits of any purchase I make.

 

Now having said all that... rarely do I stop down past f2. Given the Noct is on my body most of the time, a cleanish iso 800 is all I ask.

 

Does this correlate with most M owners 'feel' for the performance of the M sensor? Just to be clear, I test high iso in a very dark environment. I don't test high iso in good light like others.. that makes no sense. In good lighting, high iso performance, i.e 1600-3200 on the M is perfectly adequate.

[krooj]

I don't have my 240 yet, so I can't test this out for myself, but my usual "high ISO" shooting scenario happens at night and generally in a high contrast situation. I am not really concerned with shooting in dark spaces without strong point sources of light. With that in mind, an accurate light meter and exposure is more important than simply shooting whatever and manipulating the image in post to where it would have been with accurate exposure. What kind of stuff are you guys shooting that requires ISO 3200?

[huckles]

I don't have my 240 yet, so I can't test this out for myself, but my usual "high ISO" shooting scenario happens at night and generally in a high contrast situation. I am not really concerned with shooting in dark spaces without strong point sources of light. With that in mind, an accurate light meter and exposure is more important than simply shooting whatever and manipulating the image in post to where it would have been with accurate exposure. What kind of stuff are you guys shooting that requires ISO 3200?

 

I myself shoot weddings. So although 3200 isn't a heavy requirement, there are alot of times when it is too. I don't shoot totally black walls of course.. Yes there are strong light points e.t.c. But it is always good to know what the sensor can do and what latitude you have before you shoot in the field. But absolutely the key is accurate exposure. I have to disagree about us 'simply shooting whatever' and then manipulating after. I shoot to get the most optimum quality out of the sensor if the degree of post processing is better than the in-camera settings in the first place.. why wouldn't you? e.g. the M9 and ISO 640 as a max then pushing it in post... it's night and day compared to the M9 640+ in-camera.

[krooj]

I myself shoot weddings. So although 3200 isn't a heavy requirement, there are alot of times when it is too. I don't shoot totally black walls of course.. Yes there are strong light points e.t.c. But it is always good to know what the sensor can do and what latitude you have before you shoot in the field. But absolutely the key is accurate exposure. I have to disagree about us 'simply shooting whatever' and then manipulating after. I shoot to get the most optimum quality out of the sensor if the degree of post processing is better than the in-camera settings in the first place.. why wouldn't you? e.g. the M9 and ISO 640 as a max then pushing it in post... it's night and day compared to the M9 640+ in-camera.

 

Sorry about the tone of that comment - I understand that most people using something as manual as a Leica aren't likely to forget proper exposure, but there are certainly people that would try to shoot an M like a D800 on full auto, get a junk file, then muck about with it in post, albeit to no avail.

[huckles]

Sorry about the tone of that comment - I understand that most people using something as manual as a Leica aren't likely to forget proper exposure, but there are certainly people that would try to shoot an M like a D800 on full auto, get a junk file, then muck about with it in post, albeit to no avail.

 

Oh, no need to apologise! I totally agree! For me it's my job to wring out the max performance of my gear as in the enviroments I shoot, lighting is very dramatic and changes very quickly so I have to be on my toes and know what each and everypiece of equipment I have can do and of course.. what I can do. I used the D4 as a reference comparison because it is the bench mark for lowlight work. I'm very happy with the M... there are really good features on it that set it above the M9P (which I still adore)... image performance is just a tiny small up. So that could be very comforting for current M9 owners. But in terms of sensor performance, the M is between the D800 and D4 (I don't 'normalize' my images to 8Mp), but the lenses on the M of course are magnificent. and I'm in love with my Noctilux!

[JimKasson]

Hi, thought I would put my own experience with my testing, although no where near as techincal. Just observation of the M at different ISO's pushed or native vs Nikon D4 push and native.

 

Up until 800, the Nikon D4 is 'just' slightly cleaner, but the M is doing very well. At 1600+ on the M this is where the D4 streaks ahead. You can push an M 800iso file 2 stops but that's when banding occurs. It certainly does have a huge colour shift to green though. ISO 3200 native on the M is much more accurate colours (and brighter!) but exhibits the same amount of banding. Not terrible banding, but not ideal. As a comparison, you can have any ISO range in the D4 and push it 4 stops before ANY banding is visible.

 

Does this correlate with most M owners 'feel' for the performance of the M sensor? Just to be clear, I test high iso in a very dark environment. I don't test high iso in good light like others.. that makes no sense. In good lighting, high iso performance, i.e 1600-3200 on the M is perfectly adequate.

 

Your post prompted me to do some testing of the two cameras which pretty much bears out what you are saying. In order to be fair, I corrected for pixel pitch, which if uncorrected gives the D4 an unfair advantage, and ISO sensitivity, for which a lack of correction tips the scales towards the M240. Sample space of one, but My M240 is more than half a stop less sensitive to light as measured by raw values than my D4 at the same ISO, and that's way too big a difference to be explained by sample variation.

 

Graphs and details are here.

 

Jim

[huckles]

Your post prompted me to do some testing of the two cameras which pretty much bears out what you are saying. In order to be fair, I corrected for pixel pitch, which if uncorrected gives the D4 an unfair advantage, and ISO sensitivity, for which a lack of correction tips the scales towards the M240. Sample space of one, but My M240 is more than half a stop less sensitive to light as measured by raw values than my D4 at the same ISO, and that's way too big a difference to be explained by sample variation.

 

Graphs and details are here.

 

Jim

 

Hi, yep... in order to be fair, those corrections need to be done. For me though, in real world usage, I don't upscale, downscale e.t.c. I use the sensor for what the sensor is.

[JimKasson]

For me though, in real world usage, I don't upscale, downscale e.t.c. I use the sensor for what the sensor is.

 

[Thread drift warning]

 

I hear what you're saying, but don't you ever print the images? The resolution of the image needs to be scaled to match that of the printer (360 ppi/ 720 ppi for Epson, 300 ppi and maybe 600 ppi for Canon). If you don't do the scaling (or let a program like Lightroom or QImage do it for you), the printer driver will do it, and it will do it using nearest-neighbor interpolation (at least in the case of Epson), which. among other problems, doesn't reduce noise when down-sampling .

 

If you don't change resolution, you'll be printing all your M240 images at about 16 x 11 or 8 x 5.5 inches (420 x 280 or 210 x 140 mm) on Epsons or 20% bigger on Canons. Your D4 images will print about 20% smaller.

 

Jim

[huckles]

[Thread drift warning]

 

I hear what you're saying, but don't you ever print the images? The resolution of the image needs to be scaled to match that of the printer (360 ppi/ 720 ppi for Epson, 300 ppi and maybe 600 ppi for Canon). If you don't do the scaling (or let a program like Lightroom or QImage do it for you), the printer driver will do it, and it will do it using nearest-neighbor interpolation (at least in the case of Epson), which. among other problems, doesn't reduce noise when down-sampling .

 

If you don't change resolution, you'll be printing all your M240 images at about 16 x 11 or 8 x 5.5 inches (420 x 280 or 210 x 140 mm) on Epsons or 20% bigger on Canons. Your D4 images will print about 20% smaller.

 

Jim

 

Hi Jim, I certainly do print. But what I'm saying is that for me, in any useful pixel to pixel real world talk about a sensor, I like to keep it to what it gives me. When they are all printed properly, it's a much of a muchness anyway right?

How have you been going with your thoughts on the M past 800 iso?

[JimKasson]

I've conducted further testing on the green shadow issue, and have reached the following tentative conclusions:

 

 

• The M240 exhibits nonlinear response at ISOs 200, 400, and 800, artificially depressing darker tones. The darker the tone, the greater the depression. In my previous noise floor tests, I saw the noise floor to be artificially depressed at these ISOs.
  
• Adjusting the ISO setting from ISO 200 to ISO 3200 on the M240 gives more gain at each step than is correct.
  
• This is probably the cause of the green shadows, since in most lighting conditions, for most subject matter, the green channel is the strongest. Depressing all the channels nonlinearly leaves the green channel the last one standing, so to speak.
  
• Because of the above, the green channel shadow shift is probably more accurately characterized as a shadow shift in the direction of the highest-valued raw channel. I will construct a test to verify this.
  

 

Details here.

 

 

Jim

[furcafe]

Could this depression of dark tones be intended to mitigate banding?

 

I've conducted further testing on the green shadow issue, and have reached the following tentative conclusions:

 

 

• The M240 exhibits nonlinear response at ISOs 200, 400, and 800, artificially depressing darker tones. The darker the tone, the greater the depression. In my previous noise floor tests, I saw the noise floor to be artificially depressed at these ISOs.
  
• Adjusting the ISO setting from ISO 200 to ISO 3200 on the M240 gives more gain at each step than is correct.
  
• This is probably the cause of the green shadows, since in most lighting conditions, for most subject matter, the green channel is the strongest. Depressing all the channels nonlinearly leaves the green channel the last one standing, so to speak.
  
• Because of the above, the green channel shadow shift is probably more accurately characterized as a shadow shift in the direction of the highest-valued raw channel. I will construct a test to verify this.
  

http://www.kasson.com/ll/sekonic240.PNG

 

Details here.

 

 

Jim

[Advertisement]

Hallo,

Look here: Leica M

[JimKasson]

Could this depression of dark tones be intended to mitigate banding?

 

If I were trying a one-size-fits-all fix for banding, I think I'd do things that decrease dark tone contrast, not things that raise it.

 

Jim

[JimKasson]

It turns out that the M9 doesn't exhibit anywhere near as much of the green-shadows-upon-pushing issue as the M240.

 

 

 

There is something strange, however. Like the M240, the M9 errors for the darkest patches at the lowest ISOs are negative, rather than the positive errors you'd expect if dark noise is the cause of the errors.

 

 

Since the technologies of the sensors in the M9 and M240 are completely different (CMOS vs CCD), the similarity in behavior raises the possibility that the errors are the result of a deliberate design choice.

 

 

Details and graphs here.

 

 

Jim