Take this as an example: Nikkor 16mm/2.8D fisheye lens is currently supported on the Nikon D750 body but not on a similar D780.
Why DxO does not make separate “models” for cameras and lenses and then combine them automatically? Is it scientifically impossible? If yes, for what reasons exactly?
Maybe the independent models would need the type of measurements requiring equipment that is not there or simply too costly to make results as good as they currently are? Just curious, nothing really important for me. But with hundreds of new Chinese lenses the problem is becoming serious, it seems (though not for me ).
BTW, I’ve made a request for 16mm Fisheye/D780 support some time ago, with not too much hope to resolve it. Using NX Studio for those few unsupported cases.
And PLEASE, PLEASE (in vain ), if you don’t have solid technical background in the subject, PLEASE DON’T RESPOND. Perhaps it’s a topic for another forum anyway.
I remember having read, many years ago (OpticsPro times) a statement that DxO measured each lens with the available cameras. What looked like a pragmatic way to get modules with not too much effort and resources then, now looks like something that does not scale well, if at all. We might see a distributed module creation process if DxO decided to get it on the way. Adaption cost might be higher than the saved resources though. Moreover, it would break one of DxO’s key differentiators (and ball&chain too).
My speculation is, that DxO now characterises bodies and lenses separately and combines those measurements with something that would require too much power to calculate on the fly and within the scope of an ordinary app.
DxO Optics Pro 10 from 2016 provides support for 30’338 modules.
DxO PhotoLab 9.10 (now) provides support for 135’880 such modules.
With a growth of about 100k modules over those years, annual module creation needs to be 10’000 or 100 bodies paired with 100 lenses each.
I don’t get your “calculate on the fly” and “too much power” here. Why not just generate the modules offline, which given the amount of data takes milliseconds?
There must be some other reason, maybe obvious to others?
Note that in my example the bodies are very similar, Nikon D750 vs D780.
Why not group similar bodies/sensors at least?
I think DxO will have to take some effort in that direction, but I would like to know, for my own curiousity, the technical obstacles under the hood. It doesn’t sound to me like it’s all about the business; must be something technical.
I would guess that the way a lens resolves is materially affected by both the sensor and the camera system driving it. For one, a RAW file from cameras A & B are likely to be different even if they share the same sensor part. Perhaps camera A applies in-camera processing differently than camera B and this shows up more with certain lenses.
For the Nikon Z6II there are 405 lenses tested with the Nikon Z6II. And then again with an extender, a total of 817 modules.
Do I believe these are individual and physical tested? I don’t. Would be an enormous investment, storage and time consuming.
For the Z6III these figures are 412 and 412.
Since each test requires taking a picture anyway, each lens would have to be tested against a “reference camera” and reversely, each camera would have to be tested against a “reference lens”. Deciding which references to use is already questionable.
Then, translating the resulting data from a reference camera or lens to the actual device would also be a complicated process. It’s like a language translation systems : the source text is first translated to a meta language and the result is translated from the meta language to the target language. This cannot work as well as a direct translation made from the source language to the target language by someone mastering both languages (even if AI is used).
sensels get their light from different angles depending on how close to the sensor’s center they are. Some manufacturers seem to position the microlenses accordingly. This means that sensors would have to be tested with light coning straight at them, but also from left, right, above and below. At least 9 measurements, but possibly 25 or 49 or… Lighting would have to be moved close to the sensor in order to cover the whole area from all angles.
This kind of measurement would also have to be made with each set of lenses.
The total amount of data would be much greater than what is actually stored in a module that contains data in what looks like an aggregated form with reduced spatial resolution that still provides corrections within the desired property ranges for light, acutance etc.
This complicated technical setup seems to be implemented by attaching lenses to bodies and deriving differences per body for each lens and per lens for each body. This implementation requires lots of mathematical operations, but uses a pragmatic hardware setup as shown in DxO’s respective videos or here.
I agree they don’t keep copys of every camera and lense not least we are told the delay for many requests is waiting for a loan. A test is done and then matched up to new cameras and lense as they add the.
It’s unclear if DxO does that judging from this info (again) - but the pyramids were built by lots of people, over a long time and lowish labor cost. I speculate that DxO works smarter than that through modelling using an imaginary “proxy” interface that takes the square root out of manual labor.
Back to square one: Why not separate? Because DxO is not ready for it yet, also because they want to preserve the level of quality they have established technically and through sales. And, probably, because the effort is in the testing and the math that creates the module is just humming along, spitting out modules without much human assistance.
Imagine a world in which manufacturers published lens and body property data in a standardised way: No need for DxO module creation or sites like this.
).
), if you don’t have solid technical background in the subject, PLEASE DON’T RESPOND. Perhaps it’s a topic for another forum anyway.
