Brick Wall Test with strange aperture results

[thedwp]

Advertisement (gone after registration)

Had some time on my hands today so I did the most exciting thing I could think of which is to perform a brick wall test!

If this is a waste of your time please move on...otherwise I found something curious that I would be interested if anyone else has seen.

The set up:

• M11 ISO 64 / SS125
• On Tripod w/ 2 second timer
• 50mm Summilux set at 2.0
• 35 FLE set at 2.0
• Three shots each. One at 60MP, 36MP & 18MP
• I dumped the DNG's in both LRC & Capture One 22

The lenses looked nice and sharp at all sizes.

But when I looked at the lens settings (this was identical with both lenses).....at 60MP & 36MP, both LRC & C1 showed an aperture of 2.4. BUT at 18MP both showed 2.0.

I checked all my settings and there was no exp comp on.

Would be interested if anyone else has seen this or if it's strictly a software issue. The numbers showed up in the metadata as well. I won't bother posting the images because it's a brick wall.

I'd be happy to pass along the DNG's if anyone is interested.

[Photoworks]

I am sure some where in the manual it explains that the M lenses are not electronic and they don't transmit lens data.

that mean that ALL M camera estimate the F-Stop based on expose reading or inner and outer meters.

So no surprise .

[Tailwagger]

As no doubt someone will chime it before I finish writing this, the aperture is only an estimate.  There's no lens coupling so the software makes and educated guess. That said, I have noticed that the aperture estimates seem, though not in your case, to be much further off than previously with the M10/R. There it was typical to be off by about a stop or less, but with the 11, I've seen it miss by several stops.  My guess is that ether the software likely just got ported without much modification or there's a bug related to the dual amps and change of base ISO. 

[tashley]

26 minutes ago, Tailwagger said:

As no doubt someone will chime it before I finish writing this, the aperture is only an estimate.  There's no lens coupling so the software makes and educated guess. That said, I have noticed that the aperture estimates seem, though not in your case, to be much further off than previously with the M10/R. There it was typical to be off by about a stop or less, but with the 11, I've seen it miss by several stops.  My guess is that ether the software likely just got ported without much modification or there's a bug related to the dual amps and change of base ISO. 

I think I've probably had every major M release. Have I? Hmm... probably.... and I have no impression that the M11 gets it right any less often than the others in that it has always been crapshoot! 

[Tailwagger]

36 minutes ago, tashley said:

I think I've probably had every major M release. Have I? Hmm... probably.... and I have no impression that the M11 gets it right any less often than the others in that it has always been crapshoot! 

As have I, post M9.  Possibly lens, setting, metering mode, ISO dependent or all of the above. I keep pretty good track of these things. I've had with the 135 APO, for example, the camera report F22 when the lens was at F5.6.  Never seen it that far off before with any of the earlier models.   

[tashley]

32 minutes ago, Tailwagger said:

As have I, post M9.  Possibly lens, setting, metering mode, ISO dependent or all of the above. I keep pretty good track of these things. I've had with the 135 APO, for example, the camera report F22 when the lens was at F5.6.  Never seen it that far off before with any of the earlier models.   

I’ve just always assumed it was a wild guess and ignored it!

[Tailwagger]

Advertisement (gone after registration)

1 minute ago, tashley said:

I’ve just always assumed it was a wild guess and ignored it!

If so, it's gotten even wilder with the 11. 😃

[tashley]

2 minutes ago, Tailwagger said:

If so, it's gotten even wilder with the 11. 😃

If you’re looking to thread a needle I suppose the difference between passing a camel through it or a truck is moot!

[Tailwagger]

1 minute ago, tashley said:

If you’re looking to thread a needle I suppose the difference between passing a camel through it or a truck is moot!

Depends on whether it's a camel you ride or a Camel you smoke. 

[adan]

1 hour ago, Tailwagger said:

That said, I have noticed that the aperture estimates seem, though not in your case, to be much further off than previously with the M10/R. There it was typical to be off by about a stop or less, but with the 11, I've seen it miss by several stops.

Could be - or it could be that the M11 now always meters off the sensor, with a whole different angle of view (depending on lens used) - or even a variety of metering angles (spot, CW, multi-point) compared to the simpler "semi-spot" reading of the previous Ms.

For the original poster: as previously mentioned, there is no actual connection between the aperture of an M lens and the camera, electronic or mechanical. The aperture ring sits 'way out there on the front of the lens on its lonesome, with no connecting wires or circuits or levers.

The M digitals "guess" the aperture in use by comparing the brightness of the whole scene (via the little light-detector on the front, "northwest" of the Leica logo) - to the actual brightness coming through the lens and aperture.

The sensor on the front has a specific fixed view of the world, while the internal metering varies with focal length or metering pattern (for off-the sensor metering) - there can often be a mismatch in the two readings ±.

Why, you may ask?

So that the aperture can be estimated equally well with a 1932 50mm Elmar, and with a 2022 50mm APO-Summicron-M. Backwards compatability across 90 years - a core design goal of the M system.

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

 

Edited January 28, 2022 by adan

[Tailwagger]

27 minutes ago, adan said:

Could be - or it could be that the M11 now always meters off the sensor, with a whole different angle of view (depending on lens used) - or even a variety of metering angles (spot, CW, multi-point) compared to the simpler "semi-spot" reading of the previous Ms.

The change in metering methodology seemed like the most likely explanation, but I was previously unaware of how that was accomplished and the heuristic involved... thanks! Very interesting. Be very interesting to understand how it's done now.  I could imagine, from my previous life, if I were charged with implementing this it might be possible to train an AI on the various lenses over various lighting conditions and apertures. Once built, offline, it would be a pretty simple lookup in camera that would probably not take much memory or processing. Would actually be an interesting project, if I was still interested in such things. 

Edited January 28, 2022 by Tailwagger

[adan]

The real reason Leica originally included aperture estimation at all was due to the old "Italian Flag" problem. The exact pattern of red/cyan stains (or on the M8, cyan corners from add-on IR-cut filters) varied with aperture being used, per lens type. So to correct the color vignetting in the camera, the camera had to have at least a rough idea of the aperture being used, within a couple of stops (f/2 - f/5.6 - f/11).

Since they had to make that rough guess anyway for technical reasons, Leica provided it in EXIF as well.

Due to the complaints about how approximate it was - Leica deleted the aperture guess in EXIF with the first firmware version of the M10 (still used it for internal operations).

To howls of unhappiness - so they restored it in an update a year or so later. But they can't win, either way.

No-one has ever definitively revealed exactly what the "field of view" of the reference scene-sensor is. ~50mm? 35mm? 75mm?

Will a huge black 75 Noctilux be included in the reference view/measurement? Probably. Or even the old pre-1980 stovepipe 90 mm Summicrons (which come in either chrome or black). How to compensate for that - it is too old a lens to be in the menu database?

Will the reference view include the sky - even when one is using a 90 or 135 for a tight shot below the horizon?

If you photograph a building across the street, with a black or white door, and with a longer lens (picture mostly full of door) - that will fool the system in terms of aperture estimation. But in opposite directions, depending on the door brightness.

It should be noted that the external sensor plays no role at all in setting exposure - only for guessing the aperture.

Edited January 28, 2022 by adan

[Tailwagger]

[CAVEAT... at one time I understood aspects of this stuff, but I've forgotten a lot and what I did know is likely quite out of date given how fast the technology is changing. Likely there are others here that know far more than I who perhaps will chime in]

32 minutes ago, adan said:

Will a huge black 75 Noctilux be included in the reference view/measurement? Probably. Or even the old pre-1980 stovepipe 90 mm Summicrons (which come in either chrome or black). How to compensate for that - it is too old a lens to be in the menu database?

They might not have to be. The weird thing about ML is that, at least when I was dabbling in it, it's not always very well understood exactly how the algorithm is arriving at an answer (a big part of the difficulty involved when the system goes tilt). What we think is important as humans for determining a classification, might be totally ignored as irrelevant by an ML algorithm. Simplistically, the trick is achieved by training on sample data sets to find underlying patterns, often ones utterly non-obvious to the human brain, that one can assign tags to, in this case F0.95 through F22 or so. When new data arrives, its attributes are compare to the results of the training set to find the closest match (probability) and then the algorithm spits out the tag the most closely matches tag its seen in the past. It might be that specific lens data is required, but it also could be possible that by training on a sufficient number of sample photographs tagged with the aperture and focal length used (possibly not even Leica lenses) an ML algorithm could discern attributes within any arbitrary photo that indicated the aperture used.  ie. hey ML algorithm here's 100K photos taken at F0.95, a 100K at F1, F1.2, F1.4, etc. Got it? Okay now here's a photograph you've never seen before. From what you figured out from the sample data that reliably distinguishes each of these apertures from each other, take a look at this photo and tell me what aperture it was taken with. The key thing to remember is that it's the most probable answer, not necessarily the correct one in all conditions. How accurate it can be depends all sorts of stuff, not the least of which is how complete the set is and many hidden biases might exist. ie. the algorithm if not trained properly might conclude that if there's a high percentage of purple in the right hand corner of the frame, the answer is F8. That riff might be sufficiently hidden that until someone photographs their kid with a balloon at F1.4, the flaw in the training set might never surface.

What the algorithm would see as relevant in the image for providing the answer who knows. In all likelihood, we'd be surprised by the attributes it sees as crucial. But thats the magic... and irritation... of ML. 

[thedwp]

Just wanted to say thanks to all…I had no idea the aperture was a guess nor the history behind it. I guess it wasn’t a total waste of time…I’m learning something new everyday about the M, but this was very interesting.

[fabior]

I thought that the small aperture that evaluates the intensity of the light also had the task of defining the color temperature when the AWB functionality is used. But it's my idea, can anyone confirm it?

[Edax]

9 hours ago, thedwp said:

 

The lenses looked nice and sharp at all sizes.

But when I looked at the lens settings (this was identical with both lenses).....at 60MP & 36MP, both LRC & C1 showed an aperture of 2.4. BUT at 18MP both showed 2.0.

I checked all my settings and there was no exp comp on.

The interesting thing from this is that there is a different aperture detected with the M11 set to 18MP than that is detected when it is set to 36MP or 60MP. That seems a little bit strange.

The reported value is indeed often crapshoot. So it was for a reason that Leica stopped reporting it in the first firmware release(s) of the M10. Due to popular demand however, the crapshoot aperture value became available again in later firmware versions...

The question is, is this detected aperture value used internally for other purposes like aperture dependant vignetting or colour drift correction for specific lenses? Does anyone know?

[tashley]

I’d quite like the option to enter the aperture manually, as one can for coded lenses.
 

I sometimes make a written note of it if I think it will matter. When Fotos gets updated to be an ‘always on’ silent companion providing GPS assistance, a SIRI tie-in that lets you dictate aperture per frame (or at least whenever you change aperture) and then keeps it there until you verbally changed it again would be useful.
 

Then you could have two aperture tags - yours and the camera’s guess - and could later choose which to use if, from memory, you know you were being lazy that day… 

Edited January 28, 2022 by tashley

[pgk]

If you are using the metering mode which assesses subject contrast and brightness (this used to called 'matrix') then depending on the lens in use this may cause shifts in the prediction of aperture used. Such metering is a form of automation after all and there are no doubt some calculations going on when it is being used which may be altered as focal length is added into them and the resolution shift too. Second guessing exactly what the internal software does in such situations is not going to get anywhere.

You might want to try your test again with another metering mode set and also try it with the 6-bit coding off (if it is possible) to see if there are any changes. Such anomalies are one of the drawbacks of trying to integrate sophisticated electronics and fairly dumb, manual focus lenses which can only identify their model type to the camera. I'm not a fan of multi-metering ('matrix') because it assesses scenes in ways which can lead to unpredictable results.

[SrMi]

The aperture information was never correct on M cameras, and I never referred to those values. To avoid confusion, I wish the aperture values were omitted from EXIF.

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[adan]

8 hours ago, fabior said:

I thought that the small aperture that evaluates the intensity of the light also had the task of defining the color temperature when the AWB functionality is used. But it's my idea, can anyone confirm it?

The small aperture has no effect on white balance.

You can test this yourself by taking two pictures of the exact same scene using Auto White Balance - and covering the aperture with your finger for one of the exposures. The AWB will not change from one picture to the next.

The small aperture is listed in the M10 manual only as a "brightness sensor." 

It does have a second job to do, however - it changes the brightness of the viewfinder LEDS (frameline slits, and red numbers/arrows) so that they are not too bright or too dark for the scene you are pointing the camera at. Bright scene - brighter readouts........dark scene - dimmer readouts.

The camera sets WB pretty much the same way LR or Photoshop does at home on the computer hours later - analyzes the colors in the recorded image for excess yellow or blue overall, and remove that extra color. (It can be fooled by sunsets or other tricky colors, however.)